 

 

«cum...
I furl...

mmuﬁdz

0 «RPM... ‘31.... 1:.
m iuwtunﬁdg.

Suﬁ

 

 

géﬁgsﬁ? , : .. . . 7 . . _. V

n .u. 2 abunn‘lfu hi.

THESIS

CIHG GAN STATE

IIIIIIIII IIIle II

 

 

 

 

 

 

 

IIIIIIIIIIIIIIIIIIIIIII

301 688 4748

This is to certify that the

thesis entitled

SPATIAL INTERACTION
AND
THE NORTH AMERICAN FREE TRADE AGREEMENT

presented by

James J. Biles

has been accepted towards fulfillment
of the requirements for

M.A. Geography

degree in

 

 

 

 

Date August 28, 1998

 

0-7639 MS U is an Afﬁrmative Action/Equal Opportunity Institution

 

 

LIBRARY
Michigan State
, Unlverslty

 

 

 

PLACE IN RETURN BOX
to remove this checkout from your record.
TO AVOID FINES return on or before date due.

 

MTE DUE DATE DUE MTE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1/” WWW“

Chapter One

INTRODUCTION

Statement of Problem

The North American Free Trade Agreement (NAFTA) was implemented in 1994
with the objective of eliminating trade barriers; facilitating cross border movement of
goods and services; promoting conditions of fair competition; and increasing investment
opportunities among the United States, Mexico and Canada (Pastor, 1993; Huffbauer and
Schott, 1992). Even before its inception, scholars debated the potential outcomes of
NAFTA. Proponents and opponents of the agreement alike predicted that the pact would
have a profound impact on the economies of participating nations (Krugman, 1993).

Patterns of trade since 1994, however, have corroborated the assessment of a
relative minority of experts who asserted that NAFTA’s impact on the North American
economy would be negligible (Lustig et a1., 1992; Krugman, 1993). The consensus
among economists is that NAFTA has not resulted in signiﬁcant changes in the volume
of trade nor the composition of trade among participating nations (Hinojosa-Ojeda,
1996). This thesis will assess the impact of NAFTA from a geographic perspective by
examining changes in spatial interaction between the United States and Mexico since

1993. Trans-border trade between the two countries will be used to evaluate changes in

the geographic structure of trade, as well as changes in the relative accessibility of places

that generate and attract trade.

Background

Although NAFTA differs from previous free trade initiatives in joining countries
at widely different stages of economic development, the agreement advances a process of
economic integration that began more than a decade ago (Chavez and Whiteford, 1996;
Lustig et al., 1992). Following the balance of payments crisis and external debt debacle of
the early 19803, Mexico undertook a series of dramatic economic reforms, including trade
and ﬁnancial liberalization and currency devaluation. These measures culminated in the
country’s incorporation to GATT in 1986.

The rapid inﬂow of capital following Mexico’s entry to GATT was accompanied
by a tremendous increase in the volume of trade, as well as a signiﬁcant transformation in
the composition of trade. In the early 19805, manufactured goods made up less than 20
percent of Mexico’s total exports and intermediate goods accounted for approximately
one-half of its imports. By the early 19903, however, intermediate goods and
manufactured products comprised more than 80% of the total value of both Mexican
imports and exports, respectively (Hinojosa-Oj eda, 1996).

Many of the trade liberalization measures associated with NAFTA had already
been implemented by the time the agreement was ratiﬁed. In fact, in the case of Mexico,
economic reforms were much more drastic than those imposed by NAFTA. With
Mexico's admission to GATT, maximum import tariffs declined from 100 percent to 20

percent. By the time NAFTA was conceived, average US tariffs on Mexican goods had

fallen to 3.9 percent and the average Mexican tariff was about 10 percent. Following the
inception of NAFTA, average tariffs declined to an average of 1.5 percent for the United
States and 5 percent for Mexico (Pastor, 1993; Hinojosa—Ojeda, 1996).

Not coincidentally, North American trade has become increasingly concentrated
over the past decade. Since 1986 US exports to Mexico have increased three times faster
than exports to the rest of the world; by 1989 half of US trade with Latin America was
with Mexico alone. In the case of Mexico, the volume of imports and exports began to
increase rapidly following 1988, nearly quadrupling by 1996 (Hinojosa-Ojeda, 1996).

Notwithstanding the increasing volume of trade between the United States and
Mexico, the importance of each country as a trading partner varies greatly. Though the
United States is the recipient of more than 80 percent of Mexico's exports, US exports to
Mexico generally comprise only 5 to 7 percent of the country’s total exports (Pastor,
1993; Hinojosa-Oj eda, 1996).

Trade between the United States and Mexico is largely dependent on surface
modes of transportation. During the 19903 more than 90 percent of the total value of
imports and exports has been transported by truck or rail. Furthermore, trade between the
two countries is extremely concentrated. Five US states (Texas, California, Arizona,
Michigan and Illinois) regularly account for more than two-thirds of total exports to
Mexico. Likewise, ﬁve locations in Mexico (Mexico City, Chihuahua, the state of
Mexico, Baja California Norte and Tamaulipas) receive about 60 percent of all exports
from the US (Nozick, 1996).

Although Mexico is the United States' third leading trade partner, a signiﬁcant

portion of trade between the two countries involves intra-ﬁrm transactions, comprised of

raw materials, parts and intermediate goods. For instance, two commodity groups —
machinery and transportation/vehicles —- comprise almost one-half of the total value of
exports to Mexico. The vast majority of trade within these commodity groups is
composed of intermediate goods destined for assembly in Mexican maquiladora plants
and subsequent re-entry to the United States. When the movement of intermediate goods
is accounted for, the volume of US-Mexico trade is reduced substantially, perhaps by
more than 60 percent. When trade in intermediate goods is eliminated, Mexico falls to the

United States’ sixth leading trade partner (Nozick, 1996).

Assessment of NAFTA

From an economic perspective, the purpose of NAFTA is to advance an
integrative process that began with labor ﬂows following World War II and accelerated
with Mexican economic liberalization in the 1980s (Haggard, 1995). Starting with
Mexico’s Border Industrialization Program (BIP) in 1965, however, integration was
largely restricted to manufacturing activities along the US-Mexico border. The agreement
is expected to build on each participating nation's comparative advantage; make each
country's industrial base more efﬁcient and competitive; and generate substantial new
trade, investment, employment and grth (Huffbauer and Schott, 1992).

Some experts believe that high rates of growth in Mexico would expand North
American trade and spread beneﬁts throughout the entire continental economy. Potential
beneﬁts include the reduced ﬂow of undocumented immigrants and an increase in the

demand for US manufactured and luxury goods (Lustig et al., 1992).

The United States is expected to export more agricultural products and processed
foods, chemical products, capital goods, transportation equipment and automobiles.
Mexico is likely to export more fruits and vegetables, textiles and durable goods. The
Mexican economy, because it is smaller and more dependent on US trade, is expected to
gain the most; the impact of NAFTA on the US economy is not expected to be very
profound (De Janvry et al., 1997; Krugman, 1993).

From a geographic perspective, however, the agreement will promote greater
spatial interaction by improving transferability among the nations of North America,
reducing the effort (in time and costs) required to overcome distance and decreasing the
role that distance plays in impeding interaction. If its economic objectives are fulﬁlled,
NAFTA will also result in greater complementarity (expressed in terms of increased
supply and demand) among the United States, Canada and Mexico. The purported
beneﬁts of NAFTA, however, will vary among nations, regions of individual countries,
and sectors of those national and regional economies. Changes in spatial interaction as a
consequence of the agreement will impact, to varying degrees, the relative accessibility of
places that generate and attract trade (geographic structure of trade), as well as the role of
distance in impeding the movement of speciﬁc goods (commodity structure of trade) by
different modes of transportation.

Though many experts believe that NAFTA will have a positive impact for all
countries in terms of increased trade, investment, employment and growth, some scholars
contend that the beneﬁts of free trade that come about quickly in theory are realized more
slowly in reality due to differences in economic and population size and the role of

multinational corporations (the Mexican maquiladora industry).

Grinspun and Cameron (1993) assert that NAFTA will no doubt promote rapid
economic growth in certain regions of North America. However, they assert that the
regional dominance of the United States and asymmetries in power, wealth, technology
and cultural inﬂuence guarantee that beneﬁts and costs will be distributed unequally
within and among participating countries. In addition, they emphasize the importance of
other elements, such as changes in income distribution, the effects of investment ﬂows
and increased capital mobility, and differences between countries in institutional, social
and political structures in assessing the potential outcomes of the agreement.

Hanink (1994) emphasizes the relevance of both the geographic and commodity
structures of trade in assessing NAFTA. He afﬁrms that changes in the commodity
structure of trade and geographic structure of trade (spatial ﬂows of goods) result ﬁom
the potential increase in actual and perceived beneﬁts to trading partners. Trade ﬂows are
affected by place characteristics, or a country's domestic economy; however, place
characteristics are also affected by the ﬂow of goods (Hanink, 1994). Regional disparities
will likely be inﬂuenced by changes in accessibility and the role of distance in impeding
interaction. Regions endowed with better access will beneﬁt from NAFTA; declining
relative accessibility will indicate locations that are becoming less competitive.

In the context of the North American Free Trade Agreement, trans-border trade
between the United States and Mexico offers empirical evidence of the relationship
between the commodity and geographic structures of trade, as well as spatial interaction
and econorrric development. Policies such NAFTA seek to achieve economic

development, in part, by expanding trade among member nations. Trade, the movement

of goods between places on the basis of comparative advantage and factor endowments,
is a traditional focus of economic geography and spatial interaction.

In order to assess the impact of NAFTA, the role of distance in impeding trade
must be quantiﬁed. In an international context, the ﬁiction of distance that constrains
spatial interaction is comprised of two components — the impedance resulting from spatial
separation and the barriers that exist and costs that are incurred when movement takes
place across international borders (Keeble, 1985; Batten and Nijkamp, 1990).

The traditional gravity model can be applied to trans-border trade between the
United States and Mexico to evaluate changes in the friction of distance and the relative
accessibility of places in both countries. In the absence of government intervention, the
friction of distance is thought to be relatively stable over short periods of time. At least in
the short-term, then, any signiﬁcant changes in spatial interaction and the friction of
distance between the United States and Mexico can be attributed to the impact of

NAFTA.

Research Objectives

The purpose of this thesis is to establish, both theoretically and empirically, the
relationship between national policies that promote economic integration and changes in
spatial interaction. By examining this relationship explicitly, this thesis seeks to derive a
geographic basis for evaluating the impact of policies, such as NAFTA, that promote the
interdependence of national economies.

A geographic basis for policy evaluation is essential because economic analyses

typically focus on changes in the commodity structure of trade, overlooking unintended

consequences in the geographic structure of trade. In addition, though economic analyses
identify changes in the composition of trade and aggregate costs and beneﬁts, they fail to
identify changes in the relative accessibility of places that attract and generate trade and
how the process of economic integration reduces the "barrier effect" of international
borders and the friction of distance.

Changes in complementarity between the United States and Mexico following
implementation of NAFTA will be examined using binary matrices to indicate the
presence or absence of trade ﬂows between states of each country. Matrices will be
obtained to assess the overall complementarity of speciﬁc origins and destinations, as
well as complementarity in each of ten commodity groups. Row sums will indicate the
level of complementarity for each trade origin in the United States; column sums will
indicate the demand of speciﬁc commodities generated by individual Mexican states.
Matrix totals for 1994 and 1997 will be tested statistically to discern if signiﬁcant
changes have occurred in complementarity since the inception of NAFTA.

Traditional gravity models will be used to examine changes in transferability
between 1993 and 1997. Changes in the role of distance in impeding trans-border
commodity ﬂows will be evaluated by mode of transportation and commodity type. In
addition, the relative accessibility of trade origins and destinations will be quantiﬁed by
examining changes in distance decay parameters of the models. Total-constrained gravity
models are proposed to assess changes in the overall friction of distance between the two
countries. Origin and destination-speciﬁc gravity models will be used to identify changes

in the accessibility of places that generate and attract trade.

Hypotheses

In general, a greater level of interaction is expected between the United States and
Mexico following implementation of NAFTA. In addition, the overall impact of distance
in impeding spatial interaction is expected to decline as a consequence of the agreement.
However, the magnitude of distance decay parameters is expected to vary depending on
relative location, type of commodity and mode of transportation. Changes in the
impedance of distance are also likely to vary based on location, commodity type and
mode.

In general, the friction of distance of both trade origins and destinations is
expected to be greater for peripheral locations in both the United States and Mexico.
More central locations (both economically and geographically) are expected to display
less sensitivity to distance. Furthermore, intermediate goods and liberalized products are
expected to exhibit lower distance decay exponents than non-intermediate goods and
commodities subject to taxes, respectively. Although prior research suggests that rail
shipments tend to display somewhat less sensitivity to distance than commodity
movements by truck, in the case of US-Mexico trade, commodity movements by truck
are likely to exhibit lower levels of distance deterrence due to the overwhelming share of
trade that is moved by truck and large inefﬁciencies in the Mexican rail system.

NAFTA is not expected to impact places that generate and attract trade uniformly.
Because peripheral locations are expected to be more sensitive to distance, NAFTA likely
will result in the most substantial reductions in the ﬁiction of distance among

economically and geographically peripheral states in both countries.

Changes in distance decay parameters will also vary signiﬁcantly by commodity

type and mode of transportation. Since intermediate goods are generally free ﬁ'om tariffs

and already experience high levels of interaction, the greatest changes in the ﬁiction of

distance are likely to occur among non-intermediate goods. Furthermore, due to the

reasons mentioned above, liberalized goods should also display lower distance decay

exponents following NAFTA than goods that are still subject to taxes. Finally, any

reduction in the ﬁiction of distance for different modes of transportation most likely will

favor shipments by truck rather than rail.

In order to test the above hypotheses, the following formal null hypotheses are

posited in this thesis:

Total complementarity between US and Mexican states has not increased
signiﬁcantly since 1994(1).

Complementarity between the United States and Mexico in each of ten
commodity groups has not increased signiﬁcantly since 1994 (2).

No signiﬁcant decreases have occurred in the overall friction of distance
between the two countries for 1993 and 1997(3).

No signiﬁcant differences exist in the friction of distance for peripheral states
and non-peripheral states for either 1993 or 1997(4).

Changes in the distance decay exponents of peripheral states between 1993
and 1997 are no different from changes in exponents of non-peripheral states

(5).

No signiﬁcant decreases have occurred in the distance decay exponents of
intermediate goods between 1994 and 1997 (6).

No signiﬁcant decreases have occurred in the distance decay exponents of
non-intermediate goods between 1994 and 1997 (7).

No signiﬁcant decreases have occurred in the distance decay exponents of
liberalized goods between 1994 and 1997 (8).

10

No signiﬁcant decreases have occurred in the distance decay exponents of
non-liberalized goods between 1994 and 1997 (9).

No signiﬁcant decreases have occurred in the distance decay exponents of
commodity shipments by rail between 1993 and 1997 (10).

No signiﬁcant decreases have occurred in the distance decay exponents of
commodity shipments by truck between 1993 and 1997 (11).

No signiﬁcant differences exist in the friction of distance for intermediate
versus non-intermediate goods for 1994 (12).

No signiﬁcant differences exist in the ﬁiction of distance for intermediate
versus non-intermediate goods for 1997 (13).

No signiﬁcant differences exist in the sensitivity to distance of liberalized
versus non-liberalized goods for 1994 (14).

No signiﬁcant differences exist in the sensitivity to distance of liberalized
versus non-liberalized goods for 1997 (15).

Distance decay exponents for commodity shipments by truck are not
signiﬁcantly less than distance exponents for shipments by rail for 1993 (16).

Distance decay exponents for commodity shipments by truck are not
signiﬁcantly less than distance exponents for shipments by rail for 1997 (17).

ll

Chapter Two

LITERATURE REVIEW

Spatial Interaction

The term " geography as spatial interaction" was coined by E.L. Ullrnan more than
40 years ago to describe the interdependence between geographic regions; it encompasses
any movement over space that results from a human process (U llrnan, 1980; Haynes and
Fotheringham, 1984). Deﬁned as the movement of goods, people, money and
information, spatial interaction considers both site (local, underlying areal conditions)
and situation (the interrelationships between places). As a result, though spatial
interaction is clearly a function of place characteristics, it is also a product of the degree
of spatial integration within a system of places (Ullman, 1980; Hanink, 1994).

According to Ulhnan, the bases for spatial interaction are complementarity,
supply and demand considerations derived from areal differentiation, econorrries of scale
and comparative advantage; intervening opportunity, the existence of complementary
supply sources; and transferability, the costs of transportation and the effort required to
overcome the friction of distance. Spatial interaction recognizes the distorting inﬂuence
of political control and can be used to understand, or predict, potential new interaction
under changed conditions, including policies such as NAFTA that promote increased

economic interdependence (Abler et al., 1971; Ullman, 1980).

12

From a geographic perspective, spatial interaction is the process through which
economic development is achieved. The relationship between spatial interaction and
economic development is dynamic and mutually reinforcing. Economic development is
the result of increased specialization and specialization is only possible through spatial

interaction (F otheringharn and O’Kelly, 1989).

The Basic Gravity Model

The gravity model is the most widely used spatial interaction model and one of
the earliest models applied in the social sciences. Carey introduced the logical basis for
the relationship between gravity and human interaction in the second half of the 19th
century; it was ﬁrst applied by Ravenstein near the turn of the century in studying
migration between English cities (Taaffe et al., 1996; Lowe and Moryadas, 1975).
Gravity models attempt to describe the two most obvious factors affecting the amount of
ﬂow or interaction between any two locations - scale impacts (population, for example)
and distance (Taaffe et al., 1996; Haynes and F otheringham, 1984).

The gravity model assumes that two places interact with each other in direct
proportion to the product of their masses and inversely proportional to some function of
the distance between them. Interaction is deﬁned as proportional to the trip generating
capacity of the location where the trip begins and the attraction capacity of the location
where the trip ends. In addition, the volume of ﬂows decreases as distance or some other

measure of spatial separation increases:

13

(1)

where: I,-- = the amount of interaction between places i and j during
some period of time;
d; = some measure of spatial separation between places i and j
P,-, PJ= some measure of attraction between pairs of interacting
places i and j

Gravity models are particularly useful in assessing two of Ullman's bases for
interaction: complementarity and transferability. In addition, the model has been reﬁned
over the past 50 years to permit inclusion of the attributes of origins and destinations,
spatial structure, and intervening opportunity.

Since its appearance in the social sciences literature more than a century ago, the
gravity model has been used successfully to study international trade (Linnemann, 1966;
Yeates, 1969) and commodity ﬂows (Black, 1971; 1974); assess the impact of
international borders on telecommunications (Mackay, 1958; Rietveld, 1993); delineate
market area boundaries; explain migration patterns; and model urban land use (Lowry,
1964) and the urban transportation system. Though the model has been criticized for its

lack of a sound theoretical basis, its continued use results from its surprisingly high

degree of statistical explanation and empirical validation (Wilson, 1970).

Propulsive and Attractive Forces
The deﬁnition of attractive and propulsive forces (the P,- and Pj terms of the
model) depends largely on the context in which the gravity model is applied. For

example, labor surplus and number of unﬁlled job opportunities have been used in studies

14

of migration and retail ﬂoor space has been utilized in retail market potential studies. Due
to their availability, however, population and income ﬁgures (such as gross domestic
product) are the two most commonly used measures of propulsive and attractive forces.
Since population and income variables may conceal important differences between
places, they may be weighted to account for variations in population characteristics such
as income, education and gender; accommodate cultural and economic differences; and
recognize the different contribution made by different regions to interaction potential.
Propulsive and attractive forces may also be raised to some exponential power to control
for size or agglomerative effects of population or economic activity (Haggett, 1977;

Lowe and Moryadas, 1975; Desta, 1988):

P 2' P ’.
I = ’ (2)
I d U
where: 1.3-, Pi, P,- and d; are as above, and

a and A. are weights or exponents

The Potential Model

Another method of accounting for differences in propulsive and attractive forces
is the potential model. The concepts of population and market potential were pioneered
by Warntz (1959) and Harris (1954), respectively. Potential is a weighted measure of
centrality or accessibility within a set of spatial nodes and can be thought of as a measure
of the proximity of a given location to all other places in a spatial system.

The potential model is frequently used when the researcher is concerned with
ﬂows between one origin and many destinations. Whereas the gravity model is based on

the assumption that the interaction between two places is related to the product of

15

propulsive and attractive forces divided by the distance between them, the potential
model quantiﬁes interaction by summing the total interaction between a given place and

all other areas (Dicken and Lloyd, 1990; Lowe and Moryadas, 1975):

P,-

 

V ,. = (3)
Z d .-,-
where: V,- = total potential at place i
P,- = the size of another place in the region
d,j = the distance separating places i and j

When potential is computed for a large number of points in a region, a map of a
potential surface can be constructed. Potential represents a force underlying interaction
between places; like the gravity model, the basic population potential model can be
weighted by income or retail sales model depending on the focus of a given study
(Dicken and Lloyd, 1990).

The potential measure also can be used to calculate regional accessibility to
economic activity. Economic potential, achieved by weighting population potential by
income or gross state product (GSP), is a measure of accessibility to economic activity;
this model can be interpreted as the volume of economic activity a region has access to
after accounting for the cost of covering the distance to that activity. As a summary
measure, it identiﬁes disparities in accessibility between regions of high and low
economic potential. A high relative accessibility conveys a comparative advantage and
reduced distance costs. In addition, the modiﬁed potential model can be used to illustrate
how regional accessibility to economic activity has changed over time (Abler et al., 1972;

Keeble, 1980).

16

Friction of Distance

Euclidean distance, transportation costs or transportation time are most commonly
used to quantify the role of spatial separation in impeding spatial interaction. In many
instances, the distance decay term is raised to an exponential power to reﬂect the varying

negative inﬂuence of distance on interaction. This exponent is generally determined

 

empirically:
P" PA
[ij = i A J (4)
u
where: P;, P,-, d U, or and A. are as above, and

[3 is an empirically derived exponent

Although the gravity model assumes that the effect of distance varies smoothly
and continuously over space, many applications of the model have revealed a substantial
variation in the magnitude of distance decay parameters. Losch (1954), Mackay (1958),
and Rietveld (1993), for instance, have demonstrated that international borders create
signiﬁcant discontinuities in patterns of interaction among places in different political
territories.

Analyzing bank deposits along the US-Mexico border, Losch (1954) revealed that
political barriers result in the truncation of spatial interaction, producing an effect
identical to that of increasing the distance between interacting areas (Batten and Nijkamp,
1990). Though Losch does not explicitly employ the gravity model in this study, he
illustrates how political and natural boundaries, reinforced by tariffs, theoretically should

result in reduced spatial interaction.

17

Mackay (195 8), in his analysis of international and interregional telephone calls in
Ontario, Canada, demonstrates that international political boundaries create
discontinuities in the patterns of spatial interaction among places in different political
territories. By plotting distance decay parameters for several forms of movement across
borders and comparing them to values for ﬂows in a borderless environment, Mackay
also assesses the displacement in spatial interaction due to the discontinuous friction of
political boundaries. His research reinforces the utility of gravity and potential models in
assessing the discontinuities created by international borders.

Rietveld (1993), in a study of barrier effects and transportation and
communication networks in Europe, reveals that international borders exert a signiﬁcant
inﬂuence on telecommunication ﬂows. Using a modiﬁed gravity model with a “reduction
factor,” he ﬁnds that telecommunication interactions between European nations are 60 to
70 percent less than the interaction that would be expected in the absence of borders. He
attributes this reduction not to increased tariffs, but to the fact that telecommunications is
complementary to other forms of spatial interaction, such as trade and tourism.

Other studies have found that the attenuating effect of distance varies for different
kinds of movement and at different distances. As Black (1971) and Taaffe et a1. (1996)
note, the distance exponent can be expected to vary with time, different modes of
transportation, locational changes, different commodities, the areal extension of
interaction, and different degrees of regional specialization.

Numerous researchers have also noted a systematic decrease in the friction of
distance for more central locations and the tendency for steeper distance decay functions

to be found in less accessible areas (Gordon, 1985; Sheppard, 1984). Gordon (1985)

18

states that the spatial distribution of population and opportunities for interaction give rise
to variations between regions in distance decay. He asserts that real income differences,
scale economies, transport costs and the spatial concentration of specialized ﬁmctions
(the basis of hierarchies in central place theory) produce stronger distance decay effects in
peripheral locations.

Other researchers have argued that the substantial variation in distance decay
parameters is the result of model misspeciﬁcation, speciﬁcally its failure to adequately
consider spatial structure (Haynes and F otheringham, 1984). In addition to recalibrating
the model to consider the effect of competing destinations, intervening opportunities and
potential models have been suggested as means of accounting for spatial structure.
However, compelling evidence has been presented that the model is not necessarily
rrrisspeciﬁed and that including measures of spatial structure does not necessarily
improve estimates of interaction (Desta and Pigozzi, 1991). Notwithstanding the potential
for model misspeciﬁcation, Haggett (1977) asserts that comparison of relative variations
in gravity model parameter values for the same spatial conﬁgurations over time or for

different commodities, would remain valid if the spatial structure remained constant.

General Family of Gravity Models

The basic gravity model has been reﬁned to produce a general family of spatial
interaction models (Wilson, 1970). In general, four different forms of the model exist: the
total- constrained model, production-constrained model, attraction-constrained model,
and doubly constrained model. Among other factors, the choice of a model depends on

the scale of analysis, the purpose of the study and the availability of data. The doubly-

l9

constrained model is the most data intensive; it also produces the most accurate
predictions of spatial interaction. Production and attraction—constrained models require
less data and are somewhat less accurate; the total constrained-model only requires
information on the total volume of interactions in a system, but produces the least reliable

estimates of interaction.

T otal-Constrained Model

The total-constrained gravity model is used to estimate the interaction between
pairs of zones or regions when only the total volume of interaction is known. The only
model constraint is that the total volume of estimated interaction between regions equal
the overall amount of interaction. In order to ﬁrlﬁll this constraint, a balancing factor, or

scalar, is included in the model (Haynes and Fotheringham, 1984; Senior, 1979).

Production-Constrained Model

In the production-constrained gravity model, information is available on the
volume of movement ﬁom each origin in the spatial system. Obviously, the total number
of interactions in the system is known as well. The production constrained model is
useful in forecasting the total volume of interaction arriving at each destination. In
addition, this form is frequently used to calibrate origin-speciﬁc gravity models to reﬂect
the relative accessibility of origins and the perception of destination attractiveness and
distance as the determinants of interaction. A balancing factor is also incorporated to

insure that the total of predicted interactions originating in a given location equals the

20

number of actual interactions originating there (Haynes and Fotheringham, 1984; Senior,

1979).

Attraction-Constrained Gravity Model

The attraction-constrained gravity model is structurally similar to the production-
constrained model. If the total volume of interaction attracted to a system of locations is
known beforehand, the attraction-constrained model predicts the distribution of
movement from origins to destinations. The model is frequently employed in the format
of a destination-speciﬁc gravity model to assess relative differences in the accessibility of
a set of places that attract interaction. The model is also constrained by a balancing factor
so that the total estimated interaction attracted to a speciﬁc place is equal to the actual

volume of movement (Haynes and Fotheringham, 1984; Senior, 1979).

Doubly-Constrained Model

In the doubly-constrained gravity model, data are available on the total amount of
interaction leaving each origin and arriving at each destination. Actual interaction
between given pairs of origins and destinations, however, is not known and must be
estimated. In this case, constraints on origin and destination totals operate simultaneously
and interaction between a given pair of locations must be calculated iteratively (Haynes

and Fotheringham, 1984; Senior, 1979).

21

Interpretation of Model

Distance decay ﬁrnctions are generally highly skewed, with a large number of
ﬂows or contacts at short distances and a substantially less interaction at greater
distances. In addition, since spatial interaction is a function of the product of two masses,
any plot of interaction intensity values is also likely to be skewed. Such intrinsically
nonlinear relationships can be made to approximate a linear relationship through the use
of the simple log transformation. Consequently, the traditional gravity model equation (1)
may be transformed using the simple log transformation and interaction data can then be
manipulated by linear regression (Senior, 1979; Lowe and Moryadas, 1975,

Fotheringharn and O’Kelly, 1989):

log Ii,- = log a + b, log P,- + b; log P]- - b3 log dij (5)

where: 1,], P;, P,- and dij are deﬁned as above, and
a, b1, b2, b3 and b4 are estimated using linear regression
In the context of the gravity model, linear regression measures the closeness of ﬁt
between the log of actual ﬂows and estimates of that ﬂow. The coefﬁcient of
determination (r2) indicates the percentage of variation in actual spatial interaction (the
dependent variable) associated with variation in estimated interaction (independent
variables). Regression constants and parameters provide estimates of the best weighting
to apply to the negative effect of distance decay on interaction. Dummy variables and

statistical measures of signiﬁcant differences among regression coefﬁcients, such as the

22

Chow Test, can also be applied to assess variation in the friction of distance and its
change over time (Studenmund, 1992).

Though linear regression is perhaps the most commonly used form of calibrating
gravity models, other methods, including the standardized root mean square error
(SRMSE), have been suggested as alternatives when large error is present or interaction
matrices (vectors) are highly sparse (Knudsen and Fotheringham, 1986; Desta, 1988).
The SRMSE is used to calculate the most likely distance decay exponent based on actual
and expected interaction values. The distance exponent is calculated iteratively, and the
gravity model is recalibrated, until the SRMSE has been minimized and no further

reduction in error is possible:

SRMSE={ ;;(tij_t;)2/(m*n) }l/2/( 2‘: gtU/mtn) (6)

where: tij = actual ﬂow from place i to j
tif" = predicted ﬂow from place i to j
m = number of rows in interaction matrix
n = number of columns in interaction matrix

Use in International Trade Studies

The initial applications of the gravity model to international trade ﬂows were
carried out in the 19603. Linnemann (1966) developed an econometrics based model to
explain differences in the size of international trade ﬂows between pairs of countries. He
identiﬁed the three most obvious factors that were signiﬁcant in determining the volume
of trade — supply, demand and resistance. Supply and demand variables represented
attractive and propulsive forces of the model and trade resisting forces fall into two

categories — natural obstacles and artiﬁcial trade impediments.

23

Linnemann found that trade resistance varies with the type of commodity and
conﬁrmed the relevance of distance in empirical studies. He also employed a dummy
variable to account for the existence of preferential trade agreements. Linnemann
concluded that political and economic alliances may have led to the selective lowering of
tariff barriers and import restrictions and that the effects of changes in these trade
resistance factors over time, should be observable in changes in the value of the exponent
of distance.

From an explicitly geographic perspective, perhaps the most signiﬁcant
application of the gravity model to international trade was carried out by Yeates (1969).
Yeates asserted that international trade can be studied as a special form of spatial
interaction that involves analysis of movements between places that are politically
independent as well as physically separated. He employed a modiﬁed version of the
gravity model, using total national income as an operational deﬁnition of propulsive and
attractive forces and multiple regression analysis.

Like Linnemann, Yeates realized that the volume of trade between countries is
affected not only by attractive and propulsive forces such as the size or purchasing power
of countries, but also by political decisions reﬂected in multilateral and bilateral trade
agreements. He stated that the importance of these political decisions can be reﬂected in

the distance decay term.

Use in Commodity Studies
The application of the gravity model to disaggregated trade data, speciﬁcally

commodity ﬂow studies, was pioneered by Black (1971, 1972, 1974). Black (1972)

24

explicitly examined the descriptive power of gravity model applied to ﬂows of goods
between regions — interregional commodity ﬂows for the United States. Using a stepwise
procedure to calibrate the friction of distance component of production-constrained
gravity models, he discovered that the model is effective and accurate in studies where
total shipments and total demand values are known for each region. On average the
model accounted for 93 percent of the variation in commodity ﬂows. Distance decay
exponents, however, varied markedly, from 0.25 to 5.325.

Black emphasizes that the understanding of the variables that inﬂuence the
distance exponent is poorly developed. His results reinforce general economic theory,
which posits that commodity ﬂows are inﬂuenced by the cost of overcoming distance,
and the extent to which goods can bear the cost of transportation.

In a related study, Black (1971) also examined the possibility of estimating the
distance exponent using variables related to variation in the ﬁiction of distance and the
impact of change in the size of the study area. Using the same iterative process as above,
he determined that the gravity model tends to perform better with more aggregate data
and that, again, distance exponents are quite variable. He was able to identify several
variables that are related to the distance exponent — number of suppliers, the type of good,
and the degree of specialization. Black concluded that higher levels of specialization,
implying high value and a greater proportion of shipments from largest producer, result in
a lower friction of distance exponent and articulated the need to examine temporal

stability of exponents.

25

Additional applications of the gravity model to interregional commodity ﬂows
have been canied out by Chisholm and O'Sullivan (1973), O’Sullivan and Ralston
(1974), and Byler and O’Sullivan (1974).

Applying a variation of the gravity model used in the analysis of the urban
transportation system, Chisholm and O’Sullivan attempted to predict interregional
commodity ﬂows in Great Britain by mode of transportation and commodity group. In
addition, they employed the concept of economic potential as a surrogate measure of the
potential accessibility of any one place to all other places within the system and a proxy
for the facility with which the national market may be reached from each location.

Using the log-linear version of the gravity model and ordinary least-squares
regression, Chisholm and O’Sullivan conﬁrmed the hypothesis that the greater the
potential accessibility of an area, the larger its volume of trade. They also conﬁrmed that
distance exponents vary considerably from one region to another; this spatial variation is
interpreted as the result of an urban/rural duality. High values were characteristic of
peripheral areas and low values were found in more accessible, urban areas.

In support of Black’s research, Chisholm and O’Sullivan revealed signiﬁcant
variation in the friction of distance according to different commodity groups. They also
identiﬁed a tendency for more highly valued and processed goods to have low distance
decay values. Though they believe the doubly-constrained gravity model provided a
reasonably good explanation of aggregate commodity ﬂows at the national level, the
performance of the model proved less than adequate in explaining disaggregated ﬂows at

the regional level. Though the variables chosen to represent propulsive and attractive

26

forces (population and employment) were reasonably good predictors of aggregate trade,
they were inadequate for predicting volumes of individual commodity groups.

Since trade ﬂows are affected by a region’s location in the space economy,
Chisholm and O’Sullivan concluded that potential accessibility accounts for an
acceptable proportion of the variance in distance decay values. They also emphasized the
potential for research into the stability of gravity model parameters and changes over time
and, though a large portion of their interaction matrix cells were empty, they restated the
need to disaggregate trade ﬂows for each regions on a commodity-speciﬁc basis.

O'Sullivan and Ralston (1974) examined the stability of parameters of model,
especially the exponent applied to distance, over a four-year period. Using the Chow test
to determine if signiﬁcant differences existed in the coefﬁcients of multiple regression
models, their research suggested that the effect of distance should decrease over time. In
noting the potential for distance decay parameters to decline over time, O’Sullivan and
Ralston asserted that structural changes, such as improvements in technology and
government policy, can have signiﬁcant effect on the ﬁiction of distance and impact upon
its stability over time.

Byler and O'Sullivan (1974) also examined explicitly the stability of gravity
model parameters and the tendency for the friction of distance to decrease over time.
They used multiple regression to examine changes in distance and mass terms and
interpret the ﬁiction of distance coefﬁcient as the elasticity of demand for transport with
respect to distance. They concluded that a reduction in this value could be expected in

conjunction with changes in transport costs due to network or vehicle improvements.

27

Synthesis of Literature

The geographic literature cited above indicates the potential utility of the gravity
model in assessing the impact of NAFTA on US-Mexico trans-border trade. The process
of economic integration, as represented by NAFTA, is clearly encompassed in the
concept of spatial interaction and the general family of gravity models offers a means of
assessing changes in the movement of goods.

Due the dominance of the border area and the Mexico City region, the concept of
economic potential appears useful in providing an adequate measure of the attractiveness
of Mexican states as destinations of US exports. The total-constrained gravity model
allows for assessment of changes in overall spatial interaction. Production and attraction-
constrained models (origin and destination-speciﬁc models) offer the ability to identify
changes in the relative accessibility of origins and destinations.

Though the literature review identiﬁes potential difﬁculties in carrying out
analysis of disaggregated commodity ﬂow data and quantifying the impact of speciﬁc
barriers in impeding spatial interaction, it also substantiates the applicability of the
gravity model in analyzing international trade and commodity ﬂows. Studies provide
both theoretical and empirical evidence of the role of international borders in impeding
spatial interaction and the potential for decline in distance decay parameters over time
and as a result of government policy. Previous research demonstrates the availability of
methods to quantify spatial interaction and assess the signiﬁcance of changes in

interaction as a result of NAFTA.

28

Chapter Three

METHODS

Data Collection

Any analysis of changes in spatial interaction between the United States and
Mexico since the inception of NAFTA ideally should include detailed information on the
origin and destination of trade, the value and commodity classiﬁcation of goods being
shipped, and the mode of transportation. Such secondary data would be needed at the sub-
national level, preferably at the state or county scale, in order to capture meaningful
changes in spatial interaction between the two countries and possible unintended
geographic consequences. In addition, data would be required over a substantial period of
time, perhaps a decade or more, in order to account for economic reforms (such as
Mexico's ascension to GATT) undertaken during the late 1980s. Unfortunately, such
detailed trade data are not publicly available.

The Bureau of Transportation Statistics (BTS) Trans-border Surface Freight
Database is the most complete source of information on trade ﬂows between the United
States and Mexico (Nozick, 1996). This data set has been available since April 1993 and
is extracted from the US Bureau of Census Foreign Trade Statistics Program. Much of the
trade information, approximately 55 percent of the total value of US exports and 95

percent of Mexican exports, is collected electronically at the US-Mexico border.

29

The BTS database is updated monthly and provides information on the movement
of commodities between US and Mexican states. For US exports, the database provides
detailed information on the state of origin, as well as the Mexican state of destination.
Although mode data include shipments by pipeline and mail, this study will utilize data
relating exclusively to movements by truck and rail, which comprise approximately 95
percent of the value of all surface exports from the United States to Mexico. Trade is
grouped according to the two-digit Schedule B commodity classiﬁcation, based on the
international harmonized system. The database also contains details on the value and
number of shipments and the Mexican border point of entry. Shipment value refers to the
selling price of the merchandise plus insurance and freight costs to the US-Mexico
border.

Although similar data are available for US imports ﬁ'om Mexico, no information
is provided on the Mexican state of origin. As such, the database does not allow for
analysis of changes in spatial interaction of Mexican exports. Consequently, this study
will make use exclusively of US export data for the years 1993 to 1997 to assess changes
in spatial interaction between the United States and Mexico following NAFTA.
Interpretation of results will be limited to changes in overall and commodity-speciﬁc
complementarity between the United States and Mexico, as well as changes in the

accessibility of US states as sources of supply and Mexican states as sources of demand.

Data Limitations
For the purposes of this study, monthly BTS data on US exports have been

aggregated into yearly totals to account for seasonal variation in the trade of certain

30

commodities. In addition, aggregation of monthly data into annual values reduces the
number of zero cell entries in the interaction matrices used to calibrate gravity models.
Notwithstanding this adjustment, as much as 20 percent of the cells of overall US export
matrices and more than 30 percent of some commodity and mode speciﬁc matrices are
comprised of zero cell entries.

Because widespread lack of interaction may complicate model calibration, zero
values have been eliminated from analysis in several recent applications of the gravity
model to international trade ﬂows (McCallum, 1995 ; Helliwell, 1996). However, zero
cell entries are included in this study because they represent actual values that are
meaningﬁrl in assessing changes in spatial interaction as a result of NAFTA. In total-
constrained models calibrated using linear regression, a value of one will be added to zero
cell entries in order to take the log transformation. As described above, the standardized
root mean square error (SRMSE) will be employed in destination-speciﬁc gravity models
to address this problem.

Though the BTS database is limited to trans-border surface trade, lack of
information on trade by air and sea is not expected to impact signiﬁcantly on the
reliability of this study. As mentioned above, between 1993 and 1997 more than 90
percent of the value of both US exports and imports, on average, was transported by truck
and rail. Due to the meager volume and value of trade via altemate modes of
transportation, and for reasons of parsimony, all modes other than truck and railroad have
been eliminated from this study.

The BTS database is further limited due to lack of information on commodity-

speciﬁc trade ﬂows during 1993. Data were not collected at the two-digit Schedule-B

31

commodity classiﬁcation until 1994. As a consequence, all analysis involving spatial
interaction of intermediate and non-intermediate goods and liberalized and non-
liberalized goods will be limited to the period 1994 to 1997.

According to Nozick (1996), the BTS database suffers from two additional
problems — missing data and the number of records that list unknown origins and
destinations of trade. Nozick indicates that the during its ﬁrst 12 months, 5 to 10 percent
of the information on trade ﬂows was missing from the database. In addition, for the ﬁrst
year, 22 percent of the total value of exports was listed as other or unknown. As Nozick
notes, however, the quality of the database has improved substantially since changes were
made in 1994. According to BTS, the number of unknown destinations of US exports had
declined to approximately 10 percent by April 1994. Although some consideration was
given to the notion of allocating US exports lacking destination information to Mexican
states on the basis of available information and general patterns of trade, missing data
were ultimately regarded as random. Consequently, BTS database entries listed as "other"
and "unknown" have been eliminated from this study.

.Other major limitations of the database include lack of information on the origin
of Mexican exports, the use of dollar values rather than numbers of trucks, rail cars or
ﬁ'eight tonnage, and lack of a direct means of identifying maquiladora trafﬁc. Although
data on the Mexican state of origin are collected (though unpublished for reasons of
conﬁdentiality), none of these problems has been addressed, to date, by the Bureau of

Transportation Statistics.

32

Data Analysis

Hanink (1994) asserts that most policies that affect and are affected by
international trade ﬂows can be viewed as uniformly distributed across the space of
national states. This study, however, presupposes a different relationship between
national policy and geographic and economic outcomes. The impact of the North
American Free Trade Agreement will likely vary among nations, regions of individual
countries and sectors of national and regional economies. Therefore, analysis of changes
in spatial interaction and relative accessibility will focus on trade ﬂows between US and
Mexican states.

The movement of US exports to Mexico between 1993 and 1997 will be modeled
to identify changes in spatial interaction following implementation of NAFTA. In
addition to quantifying changes in the overall ﬁiction of distance between the two
countries, trade ﬂows will be modeled at the state level to reveal changes in the relative
accessibility of trade origins and destinations.

Following Knudsen (1988), binary matrices will be used to assess changes in total
and commodity-speciﬁc complementarity among origins and destinations between 1994
and 1997. Overall differences in complementarity for each country between 1994 and

1997 will evaluated statistically using the paired t-test.

Table 1. Sample Binary Matrix

 

 

 

 

 

 

 

FromlTo Destination 1 Destination 2 Destination 3 Destination 4 Total
Origin 1 1 0 0 2
Origin 2 1 1 1 O 3
Origin 3 1 1 1 1 4
Origin 4 1 0 0 0 1
Total 4 2 3 1 10

 

 

 

 

 

 

 

33

Use of Binary Matrices

As the above example (Table 1) indicates, a cell entry of one indicates the
presence of a commodity ﬂow between a speciﬁc origin and destination. A zero cell
entry, however, indicates lack of complementarity between a given pair of places. Row
totals provide a measure of complementarity for each trade origin and column sums
indicate complementarity expressed as the demand exercised by individual destinations.
Row and column totals may also be expressed as percentages in order to provide a
relative measure of complementarity and a means of quantifying changes in supply and
demand over time. For example, in the above table, the total complementarity of Origin 1
can also be expressed as 50% (2/4). Furthermore, the total number of commodity ﬂows in
the matrix (10) provides an indication of the overall level of complementarity among
origins and destinations. This quantity also may be expressed as a percentage (62.5%) to

facilitate interpretation or comparison.

Calibration of Gravity Models

As mentioned above, the BTS database only provides information on interaction
between US and Mexican states. Consequently, interaction matrices are incomplete
because no data are available on trade ﬂows that originate and terminate within the same
country (or the same state). Lack of information on ﬂows within a given state or country
may result in biased distance decay exponents as models fail to indicate how rapidly
interaction falls off with distance from the origin.

Likewise, as illustrated by the relative locations of Texas and Maine, a potential

bias may exist in the distance matrix. In no case does Texas interact with Mexican states

34

at a distance greater than 2100 miles. However, Maine interacts with no Mexican state at
a distance less than 2500 miles. In conjunction with incomplete interaction matrices, the
systematic differences in distances could result in biased distance decay exponents.

In order to compensate for these potential limitations, data ﬁom the 1993 US
Commodity Flow Survey (CFS) will be used to supplement gravity model interaction
matrices. Since the CFS only provides data on interstate and intrastate trade for US
origins, only origin-speciﬁc gravity models can be calibrated in this thesis. As a result,
analysis of changes in relative accessibility will be limited to US states that generate
commodity ﬂows to Mexican states. Although CF S data do not exist for 1997, likely
interstate and intrastate trade values can be predicted using gross state product (GSP) as
the independent variable in a simple linear regression model. Linear regression indicates
that GSP for 1993 explains 86 percent of variation in 1993 cormnodity ﬂows. By using
coefﬁcients from the 1993 regression model and GSP values for 1997, fairly reliable
estimates of 1997 commodity ﬂows can be obtained and used to calibrate origin-speciﬁc
gravity models.

BTS trade data for 1993, supplemented by the 1993 CFS, will serve to calibrate
production-constrained gravity models for a base year, representing the overall level of
spatial interaction, degree of distance deterrence, and relative accessibility of origins prior
to the commencement of NAFTA. Due to the lack of commodity-speciﬁc data for 1993,
commodity speciﬁc models will be calibrated for 1994 and 1997. Changes in the relative
accessibility of places that generate trade, as well as the ﬁiction of distance will be

calculated for each year. Distance decay parameters for 1994 and 1997 will be analyzed

35

statistically to determine if signiﬁcant changes have occurred following implementation
of NAFTA.

As Figure 1 shows, US and Mexican highway systems were used to determine the
distances between US and Mexican states. In the case of the United States, the exact
geographic centroid of each state was located. The actual point used for the calculation of
distance was determined by adjusting the position of the centroid to the nearest urban area
on the Interstate Highway System. Centroids of Mexican states were determined by
identifying the dominant urban area in each state, in addition to Mexico City. In all
instances, these urban areas were connected to the Mexican Federal Highway system.
After locating geographic centroids, the US-Mexico distance matrix used to calibrate
gravity models was obtained.

Two distinct forms of gravity models are proposed in this study. The total-
constrained gravity model will be used to examine changes in overall ﬁiction of distance
between the United States and Mexico as a result of NAFTA. The production-constrained
(origin-speciﬁc) model will be used to explore changes in the relative accessibility of
places that generate trade (US states).

Total-constrained models will be calibrated on the basis of total value of trade and
for different modes (rail and truck) and commodity groups (intermediate goods,
liberalized products, etc.). Though true attraction-constrained (destination-speciﬁc)
gravity models cannot be calibrated in this thesis, intercept and slope dummy variables
will be included in total-constrained models to determine if Mexican border states, as
centers of demand, experience a different level of distance deterrence than non-border

states.

36

 

 

 

 

 

 

 

 

 

2:39am saga 53.32 EH m: A 9:53

37

For the purposes of the total-constrained model, the overall value of exports to
Mexico will be used to represent the propulsiveness of US states (supply) and the total
value of imports from the United States will describe the attractiveness of Mexican states
(demand). As mentioned above, multiple regression analysis will be used to calibrate
total-constrained models for 1993 and 1997. The Chow Test will be used to determine if

coefﬁcients of the gravity model differ signiﬁcantly over time (Studenmund, 1992):

= (RSS,—RSSl-RSSz)/k+1
CHOW (RSS.+RSS.)/(n,—2k-—2) ‘7’

 

where: RSS1 = Residual sum of squares of pooled model
RSS1 = Residual sum of squares of model one
RSSz = Residual sum of squares of model two
k = number of independent variables
nT = number of observations in the pooled model
Origin-speciﬁc gravity models will be calibrated using the standardized root mean
square error (SRMSE). As mentioned above, these models will be used to assess changes
in the distance decay ftmctions for different origins between 1993 and 1997. In order to

assess the statistical signiﬁcance of changes in distance decay exponents of models

calibrated using the SRMSE, the paired form of the t-test will be employed.

38

Chapter Four

RESULTS

The primary hypothesis of this thesis is that implementation of the North
American Free Trade Agreement in 1994 has brought about signiﬁcant changes in spatial
interaction between the United States and Mexico. This hypothesis will be tested by
examining patterns of change in complementarity (supply and demand) and
transferability (ﬁiction of distance) from 1993 to 1997. Changes in spatial interaction will
be assessed for total trade, trade in different commodity groups and different modes of

transportation.

Complementarity

Total complementarity refers to the overall level of spatial interaction among a set
of origins and destinations. Origin-speciﬁc complementarity reﬂects a given state's
attractiveness as a source of raw materials, intermediate goods and ﬁnished products, as
well its overall level of interaction with Mexican states. Destination-speciﬁc
complementarity represents a state's level of demand of raw materials, intermediate goods
and ﬁnished products, as well as its overall level of interaction with US states.

In the context of this study, total complementarity is calculated as the sum of

interactions for a speciﬁc origin or destination in each of ten commodity groups. In order

39

to facilitate comparison of total complementarity and changes in interaction between

1993 and 1997, values have been converted to percentages.

Complementarity by State

As Tables 2 and 3 indicate, total complementarity between the United States and
Mexico increased between 1994 and 1997. Paired t-tests reveal that increases in total
complementarity are statistically signiﬁcant for both the United States (5.512) and
Mexico (2.454), signifying signiﬁcant changes in both sources of supply and demand
between 1994 and 1997. As a consequence, Hypothesis One, regarding signiﬁcance of
changes in total complementarity, is rejected.

As displayed in Figure 2, the highest levels of total complementarity among
Mexican states are found along in the border area, the Mexico City region, and the state
of Jalisco. Not surprisingly, the lowest levels of complementarity are found in southern
Mexico, the Yucatan Peninsula, and the state of Baja California Sur.

Increases in overall demand among Mexican states is concentrated in the northern
part of the country. However, substantial growth in demand can also be found among
smaller, more distant states (Tlaxcala, Morelos, Nayarit, and Campeche), as well as
central border states. Though reduced complementarity is relatively infrequent,
substantial decreases in demand are located in Durango and the border state of

Tamaulipas.

4O

Table 2. Total Complementarity, Mexico 1994-1997

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

41

MEXICO 1994 1997 Change
Aguascalientes 39.40% 46.50% 7.10%
Baja California Norte 48.50% 55.60% 7.10%
Baja Califomia Sur 11.70% 13.30% 1.60%
Chihuahua 59.80% 65.00% 5.20%
Colima 49.20% 54.20% 5.00%
Campeche 4.40% 12.30% 7.90%
Coahuila 59.80% 64.80% 5.00%
Chiapas 9.40% 8.80% -0.60%
Distrito Federal 90.40% 90.80% 0.40%
Durango 51 .70% 41 .90% -9.80%
Guerrero 13.10% 13.10% 0.00%
Guanajuato 58.50% 60.40% 1 .90%
Hidalgo 35.20% 36.70% 1 .50%
Jalisco 73.10% 76.30% 3.20%
Michoacan 35.20% 38.10% 2.90%
Morelos 40.20% 44.20% 4.00%
Mexico 81 .30% 83.10% 1 .80%
Nayarit 9.80% 1 1 .30% 1 .50%
Nuevo Leon 81.50% 83.30% 1.80%
Oaxaca 12.50% 10.00% -2.50%
Puebla 50.80% 53.50% 2.70%
Quitana Roo 13.10% 10.00% -3.10%
Queretaro 53.10% 67.90% 14.80%
Sinaloa 46.30% 43.80% -2.50%
San Luis Potosi 49.20% 55.40% 6.20%
Sonora 70.20% 73.30% 3.10%
Tabasco 1 1.50% 9.60% -1 .90%
Tlaxcala 23.50% 39.20% 1 5.70%
Tamaulipas 67.30% 57.10% -10.20%
Veracruz 37.30% 36.70% -0.60%
Yucatan 8.50% 9.20% 0.70%
Zacatecas 19.00% 23.30% 4.30%
AVERAGE 41.10% 43.40% 2.32%
T-test 1994-1997 t=2.454*

 

Figure 2. Total Complementarity, Mexico -- 1994-97

 

42

1994

[:1 4.4%m13.1%
E] 13.1%»235%
235%m4u.2%
ﬂ 402%m593%
I 598%m90.4%

1997

[:1 sa%m133%
D 133%»233%
233%m465%
465%m679%
I 67.9%m9os%

Change 1994 m 1997
[:1 .102%m-9s%
E] .9.8% m 0.0%
011% m 32%
32% m 79%
I 7.9% m15.7%

 

Among US states, Texas and California clearly display the greatest levels of
complementarity (Figure 3). With the exception of these border states, however, the
highest levels of complementarity, almost uniformly, are located east of the Mississippi
River, especially in the traditional US manufacturing belt and the Southeast. The states
least likely to supply goods to Mexico are both relatively distant from the border and
have fairly small economies. Two regions of little complementarity, in particular, are
evident in the Upper Great Plains and New England.

Like their Mexican neighbors, US states also display widespread increases in
complementarity between 1994 and 1997. Although the most dramatic increases in
interaction are scattered across several areas of the country, a general trend towards
increased complementarity can be found in areas with historically less interaction (Upper
Great Plains and Far West). Substantial gains are also noted in the Great Lakes states and
the southern part of the country. Decreases in complementarity are localized and small,
but include the states of Texas, Florida, Oregon, North Dakota, Missouri, Kentucky and

Maryland.

43

Table 3. Total Complementarity, United States 1994-1997

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

US 1 994 1997 Change

Alabama 40.90% 46.90% 6.00%
Arkansas 40.00% 41 .90% 1 .90%
Arizona 50.60% 56.60% 6.00%
California 90.00% 90.30% 0.30%
Colorado 41.60% 44.10% 2.50%
Connecticut 44.10% 51 .90% 7.80%
Delaware 22.80% 24.40% 1 .60%
Florida 63.40% 61 .90% -1 .50%
Georgia 59.40% 60.90% 1 .50%
Iowa 41.60% 43.10% 1.50%
Idaho 17.20% 20.90% 3.70%
Illinois 73.10% 73.10% 0.00%
Indiana 47.50% 54.70% 7.20%
Kansas 42.20% 45.00% 2.80%
Kentucky 41 .30% 38.80% 250%
Louisiana 45.90% 46.30% 0.40%
Massachusetts 48.1 0% 46.60% -1 .50%
Maryland 31 .60% 29.70% -1 .90%
Maine 14.10% 12.20% -1 .90%
Michigan 59.10% 64.70% 5.60%
Minnesota 51 .30% 55.30% 4.00%
Missouri 61 .90% 60.00% -1 .90%
Mississippi 35.60% 41 .60% 6.00%
Montana 5.90% 12.50% 6.60%
North Carolina 52.80% 58.10% 5.30%
North Dakota 10.00% 8.80% -1.20%
Nebraska 30.30% 34.40% 4.10%
New Hampshire 21 .30% 24.10% 2.80%
New Jersey 58.80% 60.60% 1.80%
New Mexico 26.30% 27.50% 1.20%
Nevada 18.40% 20.30% 1 .90%
New York 62.50% 63.10% 0.60%
Ohio 60.60% 63.10% 2.50%
Oklahoma 43.40% 43.40% 0.00%
Oregon 35.90% 32.50% -3.40%
Pennsylvania 59.40% 64.40% 5.00%
Rhode Island 16.30% 22.50% 6.20%
South Carolina 37.80% 44.10% 6.30%
South Dakota 15.30% 16.60% 1.30%
Tennessee 57.20% 60.00% 2.80%
Texas 97.80% 95.90% -1 .90%
Utah 27.80% 31.90% 4.10%
Mime 43.40% 45.90% 2.50%
Vermont 1 1 .30% 16.60% 5.30%
Washington 40.60% 44.70% 4.10%
Wisconsin 55.90% 58.10% 2.20%
West Virginia 13.80% 13.40% -0.40%
W oming 5.60% 9.70% 4.10%
AVERAGE 41 .10% 43.40% 2.32%
T-test 1 994-1 997 t=5.512* |

44

Figure 3. Total Complementarity, United States -- 1994-97

  
 
  
 
  
 

1994

[:1 5.6%to18.4%
[:1 l8.4%to3l.6%
315%msrs%
I 513%m73.1%
I 73.1%m9'rs%

1997

[:1 83%m209%
[:1 20.9%m344%
34.4%msr.9%
I 51.9%m73.1%
I 73.1%m9ss%

Change 1994 to 1997

E]-3.4%m-12%
D-rz%mns%
06%m280/0
I 23%msn%
I sn%mvs%

45

Complementarity by Commodity Group

Although border states are the dominant sources of supply and demand in both
countries, analysis of commodity-speciﬁc complementarity and changes in
complementarity, can be useful in illustrating varying levels of specialization in different
regions of the United States and Mexico. In general, the lowest levels of complementarity
exist in agricultural and processed food commodity groups. Not surprisingly, the greatest
degree of complementarity is found in the machinery commodity group, which is
comprised largely of intermediate goods (Figure 4).

As shown in Figure 4, two trends are apparent in complementarity between the
United States and Mexico for the period 1994 to 1997. Seven of ten commodity groups
experienced a decline in complementarity in 1995. Though these changes are not
statistically signiﬁcant, they represent an exception to the trend of greater
complementarity and are the obvious outcome of the Mexican economic crisis and
currency devaluation of December 1994. Three commodity groups, Food and Processed
Agricultural Products, Paper Products, and Textiles and Apparel, displayed increases in
complementarity in 1995 in spite of the economic downturn in Mexico.

Nine of ten commodity groups also experienced overall increases in
complementarity between 1994 and 1997. The only exception to this trend occurred in
Food and Processed Agricultural Products, which experienced peak complementarity in
1995 and has generally declined since. Table 4 reveals the statistical signiﬁcance of
changes in the complementarity of speciﬁc commodity groups. In ﬁve of ten cases,
Hypothesis Two is rejected for Mexican states. In the case of the United States,

Hypothesis Two is rejected in six of ten instances.

46

Complementarity

Figure 4. Changes in US-Mexico Complementarity, 1994-1997

 

1100

1000*-

\, x
W .m
hug...
f' -. -hm‘-.~Dn| ,
“.Vlﬂmv.“h1 . r-
4' .

700 I'

‘. ,1

 

 

 

 

 

400

 

 

 

 

1994 1995

Year

47

1996

1997

 

 

+Farm Products
"l"'PﬂXX§SedFCOd
Chem. Products
"4*"- Plastics/Rubber
+Paper Products
—.—Texrues/Apparel
"4F'INHENIWCOUOB

—Machinery

 

Vehicles/Trans.

Instruments

 

 

Table 4. Changes in Complementarity by Commodity Group, 1994-1997

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Commodity Complementarity Complementarity T-Statistlc T-Statistlc
Group 1994 1997 Mexico United States

Agricultural Products 33.0% 36.3% 2.38* 2.1 1*
Food/Processed Ag. Products 34.5% 33.5% -0.29 -0.33
Chemical Products 44.6% 46.4% 1 .2 1 .41

Rubber and Plastic Products 48.1% 51.0% 2 .18" 257*

Paper Products 39.1% 39.9% 0.50 0.69
Textiles and Apparel 33.1% 38.0% 3.12" 4.12"

Metal Products 43.3% 46.0% 1.94* 2.58""
Machinery 64.0% 66.9% 1 .84" 260*
Transportation 35.2% 37.7% 1 .68 2.39*
instruments 35.7% 37.6% 1 .33 1 .61
Agricultural Products

As Figure 5 reveals, the greatest levels of demand for agricultural products in
Mexico are found along the border and in the Mexico City region. The lowest levels of
demand are concentrated in the southern part of the country. Though Texas and
California represent the dominant supply areas within the United States, the role of the
Midwest in supplying agricultural products is readily apparent (Figure 6).

Paired t-tests reveal that statistically signiﬁcant increases occurred in the
complementarity of agricultural products for both the United States and Mexico between
1994 and 1997 (Table 4). In Mexico, the most prominent increases are concentrated in the
border area though some growth in demand is also found in the southern states of Chiapas
and Yucatan. The greatest change in the supply of agricultural products, however, is the
trend towards decreased complementarity among states in the western third of the United
States. Increased complementarity is most likely to be found in the eastern half of the

country.

48

Figure 5. Complementarity in Agricultural Products, Mexico -— 1994-97

 

49

1994

[:1 2.1%m10.4%

1:] 10.4% m 22.9%
22.9% in 433%
I 433% in 583%
I 583% m 875%

1997

[:1 2.1%m 12.5%

1:! 125%m292%
[:1 292%m4ss%
ﬂ 453%mczs%
I 625%m832%

Change 1994 m 1997
[:1 445%

[:1 44.6% 11.42%
42% to 2.1%
2.1%m 10.4%
I 10.4%1o 16.7%

Figure 6. Complementarity in Agricultural Products, US -- 1994-97

  
  
  

1994

E] 00%m9.4%
Cl 9.4%to28.l%
‘ 28.1%m46.9%
I 46.9%1065.6%
I 655%m969%

 

0.0%110125%

12.5% to 313%
313% to 53.1%
53.1% to 75.11%
753%to969%

Change 1994 to 1997

C] -18.8% to 425%

.125%m0.0%
0.0%1063%
63% to 15.6%
15.6% to 313%

IIED

Food and Processed Agricultural Products

In Mexico, the dominant sources of demand for food and processed agricultural
products are the border region, the state of Jalisco and the Mexico City area (Figure 7).
Demand of processed food products is relatively large north of Mexico City; states south
of the Valley of Mexico generally display much less complementarity. With the
exception of Texas and California, the dominant sources of supply of processed
agricultural products, like the prior agricultural commodity group, are found in the
Midwest (Figure 8).

As mentioned above, the degree of complementarity of food and processed
agricultural products actually declined between 1994 and 1997. However, some increases
in demand can be found in the border area and the Yucatan Peninsula. In general,
however, the complementarity of states in southern Mexico decreased following 1994. As
shown in Figure 8, the most signiﬁcant increases in supply among US states following
the inception of NAFTA are located in the Upper Midwest and the Upper Great Plains.
Moderate growth is also found in the Northeast and parts of the South. Decreasing
complementarity is most prominent among states in the Southwest and along the

southeast coast of the United States.

51

Figure 7. Complementarity in Food and Processed Ag. Products, Mexico -— 1994-97

 

1994

E] 2.1%512595
[:l 125%1527.1%
27.1%539595
395%5525%
I 625%t08916%

l 997

[:1 2.1%mss%
[:1 83%1515.7%
15.7%15395%

- 395% m 50.4%

I 50.4% 15 89.6%

Change 1994 in 1997

E] -102%m-9s%
[:1 -9s%1500%
E] 00%1532%
32%157.9%
I 7.9%1515.7%

Figure 8. Complementarity in Food and Processed Ag. Products, US -- 1994-97

0.11%to9A%
9.4% to 28.1%
28.1% to 438%
438%m656%
656%11196.9%

 

  
 
  
 

1997

[:1 0.0%159A%

[:1 9.4%15313%
313%m43s%
I 433%1555595
I 555%m9ss%

Change 1994 m 1997
1:] -313%m-188%
E] -l88%to-63%
-53%m011%
l 00%m53%
I 53%m155%

53

Chemical Products

A high level of demand for chemical products is found in northern Mexico,
extending from the border to the Mexico City region. In general, southern Mexico and the
state of Baja California Sur display the lowest levels of demand for chemical products
(Figure 9). With the exception of the border region, the supply of chemical products
among US states is largely concentrated in the traditional manufacturing belt and the
eastern half of the country (Figure 10).

Though moderate increases are apparent in the complementarity of chemical
products between 1994 and 1997, statistical tests do not reveal signiﬁcant changes for
either the United States or Mexico. Binary matrices reveal that complementarity of
chemical products peaked in 1996 and decreased slightly in 1997. As Figure 9 indicates,
the complementarity of Mexican border states generally declined following 1994. The
most prominent increases in demand among Mexican states are found among states
bordering the Gulf of Mexico and in a part of northern Mexico between the border and
Mexico City. States along Mexico’s Paciﬁc coast, however, generally exhibited stagnant
or declining levels of complementarity. In the case of the United States, increases in the
supply of chemical products can be found throughout the country, though a pattern of

increased complementarity is most prominent west of the Mississippi River.

54

Figure 9. Complementarity in Chemical Products, Mexico —- 1994-97

 

55

1994

|:] 42%m1a.7%
[:1 15.7%u313%
D 313%m52.1%
I 52.1%m688%
I 583%m91.7%

1997

[j 2.1%m145%
E] 145%535495
E] 354%ms0n%
I 50.0%1572995
I 72.9%593s%

Change 1994 n 1997
1:] 42594115404961
E] -lu.4%m-2.l%
[:1 4195154290
42%m104%
I 10.4%m229%

Figure 10. Complementarity in Chemical Products, US -- 1994-97

0.0%10 18.8%

18.8%11040.6%
4U.6%to563%
563% to 71.9%
71.9% to 96.9%

0.0%109A%
9.4% to 313%
313% to 53.1%
53.1% m 71.9%
71.9% to 93.8%

  
  
   
 

Change 1994 in 1997

El -155%m-125%
[:1 —125%m-3.l%
-3.l%m3.l%

I 3.1%15125%
I 125%m219%

Rubber and Plastic Products

The greatest levels of demand for rubber and plastic products is concentrated in
northern Mexico, from the border to Mexico City. Again, the lowest levels of demand are
concentrated in southern Mexican states with relatively low incomes and lower levels of
manufacturing employment (Figure l 1). Supply of this commodity group is most strongly
associated with the traditional US manufacturing core, as well as the Southeast. States in
the Great Plains and West, with the exception of California, are much less likely to
supply rubber and plastic products to Mexican states (Figure 12).

Both countries underwent statistically signiﬁcant increases in the
complementarity of rubber and plastic products between 1994 and 1997. Increases in
demand are distributed most prominently throughout central Mexico. The most
impressive increases in supply of this commodity group are scattered throughout the
United States, though increased complementarity is noted in the southern part of the
country. Relatively little change in supply is found in the northeastern part of the country,

as well as the western part of the country (Figure 12).

57

 

Figure 11. Complementarity in RuhherIPlastic Products, Mexico -- 1994-97

 

58

[:1 2.1%mrss%
I:| l8.8%m35.4%
35.4%1552.r%
52.1%1572995
I 729%m91.7%

1997

[Z] s3%mr45%
1:] 145%527.1%
27.1%msss%
583%m7sn%
I 75.0%1591.7%

Change 1994 m 1997
El 4049411545396
|:] 45395510095
00%553%
53%5145%
I 145%5229%

Figure 12. Complementarity in RubberIPlastic Products, US -- 1994-97

  
 
  
 
  
 

1 994

I: 011% m 9.4%
1:] 9.4% m 34.4%
34.4% m 53.1%
I 53.1% m 75.0%
I 750% m 95.9%

1 997

El 3.1%15250%
Cl 25.0%m438%
43.8%15525%
I 525%m813%
I 813%5100%

Change 1994 to 1997

C] 4259559495
[:1 -9.4%m-3.l%
-3.1%m3.1%
3.1%m9.4%
I 9.4%1521.9%

59

Paper Products

As Figure 13 reveals, high levels of demand for paper products are located in the
border region and the Mexico City area. States displaying signiﬁcantly lower levels of
complementarity are located in the Yucatan Peninsula, the southern part of the country
and the state of Baja California Sur. In addition to Texas and California, the greatest
levels of supply among US states are clearly concentrated in the eastern half of the
country. US states least likely to supply paper are located in the Great Plains and western
part of the country (Figure 14).

The complementarity of paper products peaked in 1995 and has declined slightly
in subsequent years. Though overall complementarity increased somewhat between 1994
and 1997, changes are not statistically signiﬁcant. Consequently, the most salient pattern
in supply and demand of paper products is the generally stagnant levels of
complementarity. As Figures 13 and 14 show, a moderate decease in complementarity
can be found in the border regions of both countries, as well as the northwestern part of
the United States. Increases in demand among Mexican states are found in Jalisco and
along Mexico’s Paciﬁc coast. Relatively few increases in supply are apparent among US
states. States displaying the most prominent increases include Michigan, Nevada and

Tennessee.

60

Figure 13. Complementarity in Paper Products, Mexico -- 1994—97

 

61

1994

[:1 2.1%514595
El 145%15354%
35.4%554295
542%555.7%
I 66.7%t089.6%

1997

[I] 2.1%1510.4%
E] 10.4%1522.9%
22.9%552.1%
52.1%1572.9%
I 72.9%593.s%

Change 1994 m 1997
E] 250%

El -250%n-42%
42951542051
42%» 125%
I 125%52511%

Figure 14. Complementarity in Paper Pro ducts, US —- 1994-97

0.0% to 125%

125% to 313%
31.3% to 50.0%
508% in 688%
688% to 100%

0.0% to 15.6%

15.6% to 34.4%
34.4% to 53.1%
53.1% to 71.9%
71.9% to 96.9%

 

  
 

Change 1994 m 1997
El 45595153495
[:1 -9.4%m-3.l%
-3.1%n.3.1%
I 3.1%15125%
I 125%521.9%

62

Textiles and Apparel

Not surprisingly, the greatest levels of demand for textiles and apparel are found
along the border and in the Mexico City region. The lowest levels of demand are
concentrated in southern and southeastern Mexico (Figure 15). The supply of this
commodity group is dominated by the US border states and the eastern part of the
country, in particular the Southeast and Northeast. In general, states west of the
Mississippi River (Upper Great Plains and Far West) have relatively little interaction with
Mexican states in this commodity group (Figure 16).

The most notable changes in US-Mexico complementarity between 1994 and
1997 are found in this commodity group. Paired t-tests reveal highly signiﬁcant increases
in complementarity for both countries. Though increasing demand is distributed fairly
uniformly throughout Mexico, the most signiﬁcant change is a dramatic decrease in the
complementarity of the border state of Tamaulipas. Among US states, increases in the
supply of textiles and apparel are most apparent in the Southeast and the manufacturing
core. The US border region, in general, exhibits a slight decline in levels of
complementarity between 1994 and 1997. In addition, though the Far West displays
relatively low levels of complementarity in this commodity group, these states tend to
exhibit either relatively little change in complementarity or moderate decreases between

1994 and 1997.

63

Figure 15. Complementarity in Textiles and Apparel, Mexico —- 1994-97

1994

[:1 2.1%1510.4%
E] 10.4%mrss%
18.8%15395%
395%15583%
I 503%1591.7%

1997

C] 2.1%15125%
E] 125%15292%
292%m47.9%
47.9%m708%
I 703%mss.4%

Change 1994 m 1997

E] 45.7%

[:1 -15.7%m00%

011%15 10.4%

10.4%15188070
I 18.8%15292%

 

Figure 16. Complementarity in Textiles and Apparel, US -- 1994-9'1r

0.0% to 125%
125% to 258%
25.0% to 375%

375% to 59.4%
59.14% in 93.8%

   
   
  
   

1997

El 00%5155%
[:1 155%5313%
313%m4c.9%
I 45.9%m7r.9%
I 71.9%1595.9%

Change 1994 in 1997

[j -9.4%m-3.1%
[:1 3.195531%
*-i='= 3.1%59.4%
I 94%1518896
I 1889611131396

 

65

Metal Products

As Figure 17 reveals, states north of Mexico City generally display high levels of
demand for metal products. Again, the greatest levels of demand are concentrated in the
border area, the state of Jalisco and the Mexico City region. Complementarity is
considerably less in the southern part of the country and the Yucatan Peninsula. Though
Texas and California display the highest levels of complementarity among US states,
Figure 18 clearly shows the concentration of supply (and production) of metal products in
the traditional manufacturing core of the United States and the Southeast. Levels of
complementarity are substantially less in the Great Plains and Far West.

Paired t-tests reveal statistically signiﬁcant increases in the supply and demand of
metal products for both countries between 1994 and 1997. In the case of Mexico, though
increased demand is found in the border region, greater complementarity also occurs in
south-central Mexico and the Yucatan Peninsula. Furthermore, though California and
Texas exhibit moderate decreases in the supply of metal products, the most prominent
growth is found in the Great Plains and the West. Figure 18, however, also reveals
moderate growth among several states in the traditional manufacturing core (New York,

Michigan, Pennsylvania, Ohio, Illinois).

66

Figure 17. Complementarity in Metal Products, Mexico -- 1994-97

1994

[:1 2.1%.» 125%

El 125%5375%
[:1 375%m542%
542%555.7%
I 66.7%t08‘75%

1997

[:1 2.1%15125%
[:1 125%m292%
[j] 292%m4ss%
I 453%1555.7%
I 55.7%15955%

Change 1994 151997
CI 425% to 40.4%
CI -10.4% 15 -42%
II] 42% m 4.2%
42%» 10.4%
I 10.4% in 18.8%

 

67

Figure 18. Complementarity in Metal Products, US -- 1994-97

  
 
  
 
  
 

1994

[:1 00%m 18.8%
[:1 18.8%r0375%
375951553195
I 53.1%1575095
I 75.0%1510094.

l 997

[:1 3.1%15 18.8%
E] 18.8%1534A%
34.4%1553.1%
I 53.1%106880/0
I 583%m9ss%

Change 1994 in 1997
[j -9.4%m-53%
[j -53%mo.0%
0.0%1563%
I 53%515595
I 155%mzs.1%

68

Machinery

Mexican demand for machinery is concentrated in northern Mexico, extending
from the border to the state of J alisco and the Mexico City area. As with most commodity
groups, the lowest levels of demand are located in the southern part of the country (Figure
19). Though the supply of machinery is greatest in California and Texas, the greatest
concentration of complementarity is found in the eastern half of the United States.
Although most US states display relatively high levels of complementarity in this
commodity group, states in the Upper Great Plains generally exhibit lower levels of
supply (Figure 20).

Statistical analysis reveals that both Mexico and the United States experienced
statistically signiﬁcant increases in the complementarity of machinery between 1994 and
1997. Due to very high existing levels of spatial interaction, however, the
complementarity of core regions in both counties failed to undergo signiﬁcant growth.
The greatest increases in Mexican demand were found in non-border states in the
northern part of the country and Carnpeche, in the Yucatan Peninsula. Though the most
substantial increases in supply among US states are concentrated in the Far West and the
Deep South, to a lesser degree, the most prominent change in complementarity is the
clear pattern of decreasing interaction among states in the Midwest and Upper Great

Plains.

69

Figure 19. Complementarity in Machinery, Mexico -- 1994-97

 

70

1994

[:1 83% m 333%

1:] 333% m 52.1%
52.1% n. 70.8%
708% m 85.4%
I 85.4% m 97.9%

1997

[:1 145%1525.0%
1:] 250%15395%
E] 395%570895
7089610851496
I 85.4%151009’0

Change 1994 m 1997
E] -125%m-83%
[:l -83%100.0%
[:1 00%1553%
53951» 145%
I 145%1527.1%

Figure 20. Complementarity in Machinery, US -- 1994-97

1994

1:] 125%528.1%
[j 28.1%155319’0
53.1%15719%
I 71.9%1584.4%
I 84.4%10100%

  
 
  
 
  
 

1997

[:J 125%

[:1 125%537595
375%5525%
I 625%h78.l%
I 78.1%t01000/o

Change 1994 m 1997
1:] -9.4%m-53%
[:1 -53%150.0%
00%1553%
53%5125%
I 125%m188%

71

Transportation Products

In general, border states, the Mexico City area and the state of Jalisco represent
the dominant centers of demand for vehicles and transportation products. As Figure 21
indicates, demand is relatively low among states in southern Mexico, from Guerrero
along the Paciﬁc coast to the Yucatan Peninsula. In addition to Texas and California,
Figure 22 reveals that the supply of transportation products is clearly centered on the
traditional US manufacturing core, the Great Lakes and parts of the Midwest. With the
exception of border states, the western part of the country displays relatively little
complementarity in this commodity group.

Though statistical tests indicate signiﬁcant changes in the supply of transportation
products for US states between 1994 and 1997, increases in the demand for transportation
products among Mexican states are not signiﬁcant. This result is not surprising since this
commodity group largely represents intermediate goods that originate in a number of US
states destined for assembly in Mexican border states and eventual re-export. With the
exception of the state of Campeche in the Yucatan Peninsula, the greatest increases in
demand are found along the border and in the northern part of the country. In the case of
the United States, supply increases are fairly widespread, including the Upper Great

Plains, parts of the South and the East Coast.

72

Figure 21. Complementarity in Transportation Products, Mexico —- 1994-9'1r

1994

[:1 2.1%m 10.4%

[:1 10.4%m16.7%
15.7%m4r.7%
41.7%5525%
I 625%t091.7%

 

1997

E] 42%5125%

|:] 125%1527.1%
27.1%5438%
43.8%5525%
I 525%587595

   

Change 1994 to 1997
1:1 425% to -10.4%'
E1404%m-2.1%
--2.1%154.2%
ii? 42%m104%
I 10.4%1522.9%

 

73

Figure 22. Complementarity in Transportation Products, US -- 1994-97

0.0% to 125%

125% to 313%
313% to 438%
438% to 59.4%
59.4% to 96.9%

   
  
   
  
   

1997

El 3.1%m125%

III 125%1525.0%
25.0%1043896
I 438%15555%
I 555%1593895

Change 1994 to 1997

E] 455% 10 9.4%
9.4% n. 3.1%
-3.1% n. 011%

0.0% 110 63%
63% 15183020

lien

74

Instruments

Among Mexican states, the greatest levels of demand for instruments are found in
the border area, J alisco and Mexico City (Figure 23). Very low levels of complementarity
are apparent among states in southern and southeastern Mexico. With the exception of
California and Texas, the highest levels of complementarity in this commodity group are
concentrated in the northeastern quarter of the country, including the Great Lakes states
and parts of the East and Midwest. Though the South displays relatively little
complementarity, the lowest levels of supply are found in the Upper Great Plains and the
western part of the United States (Figure 24).

Though moderate increases in the complementarity of instruments are apparent
for both the United States and Mexico between 1994 and 1997, paired t-tests reveal that
changes are not statistically signiﬁcant. The most prominent changes in demand among
Mexican states are large decreases along Paciﬁc and Atlantic coats, as well as moderate
increases in the center of the country and in the border states of Chihuahua and Coahuila.
Among US states, fairly large decreases are found in the West and the Upper Great
Plains. Moderate increases are identiﬁed in the Great Plains and the South, as well as

parts of the Northeast.

75

Figure 23. Complementarity in Instruments, Mexico -- 1994-97

1994

[:| 42951512595
[:1 125%t0258%
25.0%1541.7%
ﬂ 41.7%5583%

I 583% to 875%

1997

C] 2.1%5145%
[:1 145%1527.1%
27.1%1535.4%
I 35.4%15525%
I 525%1591.7%

 

Change 1994 m 1997

CI 45.7% m 53%
[:I 53% 15 00%
0.0%1553%
I 53%15125%
I 1259680208941

 

76

Figure 24. Complementarity in Instruments, US -- 1994-97

  
 
  
 
  
 

1994

E] 00%15125%
|:] 125%528.1%
28.1%1546.9%
I 45.9%1571.9%
I 71.9%5100%

1997

[:1 00%m125%
[I] 1250411531395
313%1550.0%
I 500%1571.9%
I 719%m933%

Change 1994 in 1997

[:I -2rs%m-125%
[:1 4259553185
-3.1%ms.1%
3.1%.. 125%
I 125%5525095

 

77

Assessment of Changes in Complementarity

Although binary matrices indicate that signiﬁcant increases have taken place in
the complementarity of both the United States and Mexico, no dramatic changes have
occurred in the sources of supply and demand between 1994 and 1997. As maps indicate,
border regions and core areas of both countries generally retained their dominant role as
centers of supply and demand.

In general, the most prominent increases in overall complementarity among
Mexican states are concentrated along the border and the area north of Mexico City.
Notwithstanding the continued dominance of these regions, the most surprising changes
in complementarity following 1993 include the substantial increase in demand of
Campeche in the Yucatan Peninsula and the dramatic decrease in the complementarity of
the border state of Tamaulipas. Whereas increases in demand in Campeche may be
attributable to recent government attempts to exploit off-shore oil deposits in the Gulf of
Mexico, no explanation of the sudden decline in demand in Tamaulipas is readily
available. Changes, however, may be indicative of a greater degree of specialization in
the state’s economy or, less likely, a decline in competitiveness.

In the case of the United States, increased complementarity with Mexican states
appears to be more widespread, although clear increases are apparent among states with
historically less interaction with Mexican states. Though the most notable clustering of
increased interaction is located in the western part of the country, moderate increases are
also found in the South and parts of the Great Plains and Midwest.

Binary matrices and analysis of commodity-speciﬁc complementarity also

illustrates differing levels of regional specialization and the complementary nature of

78

interaction among certain commodity groups. Examples include similarities in the
locational patterns of agricultural and food supply centers (centered on the Midwest) and
the concentration of machinery, metals and transportation products in the traditional US
manufacturing core. Among Mexican states a zone of high complementarity in most

commodities extends from the border region to the Mexico City area.

Origin-Speciﬁc Gravity Models

Due to the potential bias in gravity models cited above, only origin-speciﬁc
models could be calibrated for 1993 and 1997 using BTS data supplemented by the 1993
US Commodity Flow Survey.

As the following maps reveal (Figure 25), the lowest distance decay exponents are
concentrated in the eastern half of the United States, as well as California and Nevada.
Moderate values are found along the border (with the exception of New Mexico) and the
Midwest. The largest distance decay exponents, by far, are clustered in the Upper Great
Plains.

The results of the origin-speciﬁc models coincide with the relationship proposed
in Hypothesis Four. In general, states with small economies that are distant from the US-
Mexico border exhibits greater sensitivity to distance. States with larger economies
display smaller distance decay exponents. Though minor exceptions to this pattern can be
found in New England and Nevada, Hypothesis Four is generally rejected.
Geographically and economically peripheral states tend to have larger distance decay

exponents.

79

Figure 25. Origin- Speciﬁc Distance Decay Exponents, 1994-97

  
 
  
 
  
 

80

1993

III 0.78451255
lj 125551.798
1.79852281
I 228153.158
I 315853.971

1997

[:1 0.75951257
[:1 125751.942
194252.477
I 257753.154
I 315453.917

Change 1993 5 1997

1:] .0147

[:1 .0147 5 -0.038
-0038 5 -0.008
I -0008501112
I 051250033

As Figure 25 indicates, however, the most substantial decreases in distance decay
exponents between 1993 and 1997 are found in the western half of the country,
particularly among states with the largest distance decay exponents. Although some
scattered decreases are found, distance exponents remained relatively stable or increased
slightly in the Midwest and East. In general, however, distance decay values decreased
(from an average of 2.089 in 1993 to 2.078 in 1997) and paired t-tests indicate that the
moderate decline in the friction of distance during this period of time is statistically
signiﬁcant (2.686) at a signiﬁcance level of 0.95. Therefore, Hypothesis Five is rejected.
The greatest decreases in distance decay exponents occur among peripheral states rather
than non-peripheral states.

As the literature review above indicates, distance decay exponents generally are
thought to be stable over relatively short periods of time. In order to determine if changes
in distance decay exponents between 1993 and 1997 are attributable to changes in trade
patterns between the United States and Mexico or ﬂows within the United States, a
second origin-speciﬁc model was calibrated using CFS data for US states exclusively as
origins and destinations. If no signiﬁcant changes are identiﬁed in the distance decay
exponents of this second model, then changes can be attributed exclusively to changes in
trade ﬂows between US and Mexican states.

Gravity models calibrated using the SRMSE reveal a slightly larger average
distance decay exponents (2.092 for 1993 and 2.082 for 1997). Paired t-tests (2.229)
indicate that this decrease in the friction of distance between 1993 and 1997 is
statistically signiﬁcant. Therefore, at least in these origin-speciﬁc models, signiﬁcant

decreases in the role of distance in impeding spatial interaction between the United States

81

and Mexico between 1993 and 1997 cannot be attributed exclusively to the inception of

NAFTA.

T otal-Constrained Models

Total-constrained gravity models were calibrated using linear regression in order
to examine overall changes in the ﬁiction of distance, as well as changes in total trade,
trade by mode and trade by different commodity groups. Because the origin-speciﬁc
models above provide some indication of changes in relative accessibility of US states as
sources of supply, intercept and slope dummy variables have been included in total-
constrained models to determine if Mexican border states experience a different level of
distance deterrence. In general, border states are expected to exhibit less sensitivity to
distance than non-border states. Therefore, the total-constrained models calibrated in this
section will quantify changes in the overall ﬁiction of distance between the two countries,
as well as differences in distance decay for Mexican border states. As mentioned above,
the Chow test will be used to identify signiﬁcant changes in the coefﬁcients of regression
models.

As Table 5 indicates, gravity models calibrated using linear regression provide
only moderate goodness of ﬁt. In general, supply, demand, distance and the dummy
variables mentioned above explain less than 60 percent of the variation in trade ﬂows. In
all regression models a signiﬁcance level of 0.95 has been chosen. Since the Chow test is
an application of the F-test, a minimum value of 3.17 is needed (with 5 degrees of
freedom in the numerator and greater than 120 degrees of freedom in the denominator) in

order to reject the null hypothesis at a signiﬁcance level of 0.99.

82

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

000.0 R50 08.50000 + 5205555 - 80805.02; - 005200.500; + 20005.55; + 55. $.- 00 808 03.82
000.0 05.0 08.00000 + 50.000000 - 800205.590 - 000250.000; + 0850.000; + .0000- 00 0085 0.5-002
000.0 055.0 500.5003 - 50550005 + 000805.000; - 0005090.? + 00050.05: + .0300- N0 00006 .0:
000.0 555.0 00250000 + 5055 000.0 - 800205.500; - 000550.500; + 20000.05: + .0000- 00 88o .0:
000.0 005.0 08.00500 + 50550505 - 80955.3: - 000500.500 + 00000.00: + .555? N0 0008 05-002
000.0 0 00.0 08.0.33 + 501.00.000.00 - 80855.55, - 000500.500 + 0.800.000; + .0000? 00 0080 05-002
000.0 05.0 5025550; + 50.05.000.00 - 000805.050 - 000500. 50.0 + 0800.003 + .5000- .00 0008 .00
000.0 005.0 82555.0 - 50.000000 + 800205.000; - 000500.00; + 20000.00: + 000. :8 00 0086 as
000.0 00.0 08.0.5.0 - 50.05.000.00 + 800205.000; - 000509.00 + 5800.050; + 050.0.- 00001
000.0 005.0 08.0.0000 - 5055.03.50 + 85.05.0000 - 000550.00: + 00000-000; + .8050. 00 00.5
000.0 005.0 82030.0 + 5050.00.00 - 800005.090 - 052593.: + 00000.03; + 050.00- 00 «8:
000.0 505.0 08.5.35. I 50.05.0053 - 000805.000 - 000550.500; + 0.850.005; + 5.00.00- 00 000:
000.0 005.0 0025000; + 50.05.505.0e - 8:805 .050; - 000500.00: + 0800-0053 + 0:00- 00 000: 55
000.0 005.0 5025.00: + 5055.00.00 - 808050000 - 000500.00; + 0.805.000; + .0000- 00 000: EB
no.5 0m ..€< :ozazum Enos.

 

 

 

 

 

£33: 6.330 V8§~§§0§U4§3h \e 002.08% .m 030,—.

83

Total Trade

Linear regression models indicate that the overall friction of distance between the
United States and Mexico declined from 2.242 in 1993 to 1.953 in 1997. In addition,
intercept dummy variables are signiﬁcant for border states for both years. Though border
states display a signiﬁcantly lower sensitivity to distance in 1993 (0.809), distance decay
coefﬁcients for border states were not signiﬁcantly different ﬁom overall distance values
in 1997. The Chow test (17.89) indicates that changes in distance coefﬁcients of
regression models for 1993 and 1997 are statistically signiﬁcant. Therefore, Hypothesis

Three is rejected. Signiﬁcant changes in the overall friction of distance occurred between

1993 and 1997.

Table 6. Signiﬁcance of Changes in Distance Decay Exponents

 

 

 

 

 

 

 

 

 

 

 

 

Model Distance Exponent Distance Exponent Signiﬁcance
1993 1997 (Chow)
Total Trade 2242* 1.953“ 1789*
Truck 2373* 2.166“ 15.2"'
Rail 0966* 1.666 23.08“
Intermediate Goods 1033* 2.016 563*
Non-Intermediate Goods 1.745“ 1.714“ 1073*
Liberalized Goods 1885* 1.277* 3.57“
Non-Liberalized Goods 2.125“ 1.709" 1059*

 

Trade by Mode

Truck Trade

 

Total-constrained models indicate that the friction of distance for truck trade
declined from 2.373 to 2.166 between 1993 and 1997. Not surprisingly, since truck trade
comprises more than 90 percent of total trade modeled above, intercept dummies are

signiﬁcant for both years and the 1993 slope dummy is signiﬁcant. Models indicate that

84

border states experienced a signiﬁcantly lesser friction of distance (0.759) for truck trade
than non-border states in 1993. However, as was the case with total trade, no signiﬁcant
differences existed in distance decay exponents for 1997. The Chow test (15.2) reveals
that changes in distance decay coefﬁcients between 1993 and 1997 are statistically
signiﬁcant. Therefore, Hypothesis Eleven is rejected. Signiﬁcant decreases have occurred

in the friction of distance for shipments by truck since 1993.

Trade by Rail

Because of the relatively sparse rail matrices and small volumes moved by rail,
the goodness of ﬁt of linear regression models for commodity shipments by rail are
substantially weaker than those of other models. As a consequence, model results are
somewhat more difﬁcult to interpret. Models reveal that the ﬁiction of distance for trade
by rail increased from 0.966 in 1993 to 1.666 in 1997. In addition, both 1993 and 1997
models indicate the Mexican border states are more sensitive to distance than non-border
locations. As Table 5 shows, the distance decay exponent for border states in 1993
(3.248) increased markedly by 1997 (4.176), indicating an increasing degree of distance
deterrence for rail shipments among border locations.

Although rail shipments generally appear to experience somewhat less distance
deterrence than shipments by truck for both years, the model indicates that border states
continue to experience a greater sensitivity to distance and that the ﬁiction of distance for
rail shipments, among border states and non-border states alike, increased substantially
between 1993 and 1997. The Chow test (23.08) conﬁrms that increases in distance decay

coefﬁcients between 1993 and 1997 are statistically signiﬁcant. As a consequence,

85

though signiﬁcant changes have occurred in the magnitude of distance decay parameters,
Hypothesis Ten cannot be rejected. Rail shipments have become increasingly sensitive to

distance since 1993.

Truck Trade vs. Rail Trade

For 1993 trade by truck exhibited an overall distance decay exponent of 2.373,
whereas distance decay exponents for rail shipments are signiﬁcantly lower (0.966). By
1997, however, the distance decay coefﬁcient for truck trade had declined to 2.166
whereas shipments by rail, though still less sensitive to distance than trade by truck, had
increased substantially (1.666). In the case of interaction with the US border, a different
relationship was apparent. In general, Mexican border states exhibited dramatically less
sensitivity to distance in the case of truck trade than movement by rail for both years.

Though the Chow test reveals that differences in distance decay coefﬁcients for
trade by mode are statistically signiﬁcant for both 1993 (168.19) and 1997 (27.98), null
Hypothesis Sixteen and Hypothesis Seventeen cannot be rejected. The Chow test indicates
that rail, rather than truck shipments, are less sensitive to the impact of distance.
Although rail shipments are becoming increasing sensitive to the ﬁiction of distance,
their distance decay exponents are still signiﬁcantly less than those for shipments by

truck.

86

Trade by Commodity Type
Intermediate Goods

Surprisingly, total-constrained models indicate that the ﬁiction of distance for
intermediate goods increased markedly between 1994 (1.033) and 1997 (2.016).
However, models also reveal that no signiﬁcant differences existed in the distance decay
exponents of border states for either 1994 or 1997. Although the Chow test (5.63)
indicates that increases in distance decay parameters are statistically signiﬁcant,
Hypothesis Six cannot be rejected. No signiﬁcant decreases occurred in the distance decay

exponents of intermediate goods between 1994 and 1997.

Non-Intermediate Goods

Gravity models reveal that the role of distance in impeding the movement of non-
intermediate goods decreased slightly between 1994 (1.745) and 1997 (1.714). Intercept
and slope dummy variables are signiﬁcant for Mexican border states in 1994; the slope
dummy variable indicates that the friction of distance for non-intermediate goods among
border states was very low (0.271) in 1994. However, no statistical signiﬁcance was
found in slope dummy variables for 1997, indicating that border states did not experience
signiﬁcant differences in the ﬁiction of distance following implementation of NAFTA.
Results of the Chow test (10.73) reveal that changes in regression model coefﬁcients are
statistically signiﬁcant. Therefore, Hypothesis Seven is rejected. A signiﬁcant decrease
occurred in the distance decay exponents of non-intermediate goods between 1994 and

1997.

87

Intermediate Goods vs. Non-Intermediate Goods

Total-constrained gravity models suggest that intermediate goods displayed less
sensitivity to the ﬁiction of distance than non-intermediate goods in 1993. However,
models also indicate that intermediate goods had become more sensitive to the impact of
distance than non-intermediate goods by 1997. The Chow test reveals that differences in
distance decay exponents among intermediate and non-intermediate goods were
statistically signiﬁcant for both 1994 (8.27) and 1997 (10.96). Data analysis indicates that
intermediate goods became more sensitive to distance than non-intermediate goods
between 1994 and 1997. As a result, Hypothesis Twelve and Hypothesis Thirteen are

rejected.

Liberalized Goods

Multiple regression models indicate that a substantial decrease took place in the
role of distance in impeding the movement of liberalized. goods between 1994 and 1997.
Distance decay coefﬁcients declined from 1.885 in 1994 to 1.277 in 1997. Models also
reveal that border states did not experience signiﬁcantly different distance decay
functions in either 1994 or 1997. Since the Chow test (3.57) indicates that the decrease in
regression model coefﬁcients between 1994 and 1997 is statistically signiﬁcant,
Hypothesis Eight is rejected. A signiﬁcant decline took place in the value of distance

decay parameters for liberalized goods between 1994 and 1997.

88

Non—liberalized Goods

Total-constrained models also reveal substantial reductions in the ﬁiction of
distance among non-liberalized goods between 1994 and 1997. Models indicate that the
overall friction of distance declined from 2.125 in 1994 to 1.709 in 1997. As in the case
of liberalized goods, border states did not display signiﬁcantly different distance decay
exponents for either 1994 or 1997. The Chow test (10.59) indicates that the decrease in
distance deterrence is statistically signiﬁcant. Therefore, Hypothesis Nine is rejected.
Non-liberalized goods became signiﬁcantly less sensitive to the friction of distance

between 1994 and 1997.

Liberalized Goods vs. Non-liberalized Goods

Regression models suggest that liberalized goods were less sensitive to the
impedance of distance than non-liberalized goods for both 1994 and 1997. In addition,
models suggest that changes in the distance decay exponents of liberalized goods between
1994 and 1997 were greater than those in non-liberalized goods. The Chow test reveals
that regression coefﬁcients were signiﬁcantly different in 1994 (17.81). In addition,
signiﬁcant differences in distance decay exponents persisted in 1997 (27.36). Therefore,
Hypothesis Fourteen and Hypothesis Fifteen are rejected. Liberalized goods displayed

signiﬁcantly less sensitivity to distance in both 1994 and 1997.

Assessment of Gravity Models
In general, origin-speciﬁc gravity models corroborate the results of the analysis of

complementarity above. Models conﬁrm that core states, in general, have retained their

89

dominance as the most accessible locations in the United States. Models also clearly
illustrate the moderate improvement in the accessibility of peripheral locations in the
United States following 1993. In addition, the relative stability in distance decay
exponents among US states in the Midwest and the Northeast serves to reinforce the
relative stability in complementarity among these locations.

Although origin-speciﬁc models conﬁrm that a signiﬁcant decrease occurred in
the overall friction of distance between 1993 and 1997, they suggest that changes may not
be due to the impact of NAFTA exclusively. Total-constrained models, calibrated with
data for US origins and Mexican destinations exclusively, also reveal that signiﬁcant
changes have taken place. Though total-constrained models provide no indication of
changes in the relative accessibility of origins, they do suggest that the overall ﬁiction of
distance between the United States and Mexico declined after 1993. Though changes in
distance decay cannot be attributed directly to the impact of NAFTA, models clearly
conﬁrm a pattern of decreasing ﬁiction of distance following implementation of the
agreement.

The relatively weak goodness of ﬁt of regression models, however, indicates that
factors other than supply, demand and distance are responsible for a signiﬁcant amount of
variation in trade ﬂows between different regions of the United States and Mexico.
Possible factors include the role played by multinational corporations and maquiladoras,
historical linkages between US suppliers and Mexican companies, the corporate structure
and locational decisions of both US and Mexican businesses, and the demand for

different commodities exerted by Mexican government agencies.

90

The results of total-constrained models generally conform to expected outcomes
in the case of total trade, shipments by truck, non-intermediate, liberalized, and non-
liberalized goods. However, the results of models for rail shipments and intermediate
goods were particularly surprising. In addition, the inclusion of intercept and slope
dummy variables for border states proved somewhat disappointing.

In most instances, dummy variables do not indicate signiﬁcant differences in
distance decay functions among border and non-border states. Interestingly, however,
slope dummy variables for total trade, though signiﬁcant 1993, are not signiﬁcant for
1997. Therefore, total-constrained models indicate that border states did not experience
signiﬁcantly different distance decay exponents than non-border states following
implementation of NAFTA. In other words, the relative accessibility of non—border states
appears to have improved somewhat after 1993.

The generally poor performance of dummy variables may have resulted from the
inclusion of the Mexico City region and the state of Jalisco with peripheral states in
southern and southeastern Mexico. The tremendous diversity in the economies of these
states probably does not permit reliable assessment of actual differences in distance decay
exponents.

The most surprising outcome of total-constrained gravity models includes the
tremendous increase in the ﬁiction of distance for rail shipments following 1993. Though
rail transportation still proves less sensitive to distance than truck shipments, increases
following 1993 may be the result of attempts by the Mexican government to privatize the

country’s rail system, ongoing government disinvestment and the increasing unreliability

91

of the system. The extreme values of border states may be an indication of the relatively
short hauls of certain commodities among US and Mexican border locations.

The dramatic increase in distance decay exponents for intermediate goods was
also unexpected. This result suggests that the movement of these products has become
increasingly concentrated in the border areas of both countries. This outcome is
especially unexpected; following the inception of NAFTA, most experts expected that
maquiladoras, and the resulting ﬂow of intermediate goods, would become more

dispersed throughout Mexico.

92

Chapter Five

INTERPRETATION AND CONCLUSIONS

Despite data limitations, this thesis has been moderately successful in assessing
changes in the geographic structure of trade following implementation of the North
American Free Trade Agreement. Although the gravity models calibrated in this thesis do
not represent truly causal models, they do provide an indication of changes in spatial
interaction, as well as the accessibility of places that generate and attract trade.

As the previous chapter reveals, border and core areas of both the United States
and Mexico have retained their dominant roles as centers of supply and demand since the
inception of NAFTA. However, results also suggest that the beneﬁts of free trade, at least
in a geographic sense, have been distributed differently, and unequally, in each country.

In the case of Mexico, binary matrices indicate that total complementarity has
increased signiﬁcantly since 1994 (Figure 2). In a spatial sense, more than two-thirds of
Mexican states exhibited greater levels of demand in 1997. However, increased
complementarity is clearly concentrated in the northern part of the country, from the US-
Mexico border to the Mexico City area. An overall decline in interaction is apparent in
the economically and geographically peripheral states of southern and southeastern

Mexico.

93

Among US states, growth in complementarity is somewhat more widespread than
in Mexico (Figure 3). Ahnost 80% of states displayed greater levels of supply in 1997
than in 1994. In addition, increases can be found in every region of the country.
Decreases are relatively few, small and geographically dispersed. Origin-speciﬁc gravity
models corroborate the results of binary matrices (Figure 25). Following implementation
of NAFTA, most states displayed relatively less sensitivity to the ﬁiction of distance.
However, US states displaying generally high levels of interaction with Mexico
experienced little change in complementarity and magnitude of distance decay exponents.
The most prominent improvements in total complementarity and transferability were
experienced by states that historically have had less interaction with Mexico.

As Chapter One afﬁrms, geographic differences in the distribution of beneﬁts of
NAFTA can be attributed to disparities in the overall economic development of Mexico
and the United States and regional economic structures within each country. As a
consequence, the outcomes identiﬁed above may be better understood in the context of
two complementary frameworks -- Myrdal's (1957) theory of circular and cumulative
causation and the “new” trade theory pioneered by Krugrnan (1994, 1995), among others.

Myrdal's theory of circular and cumulative causation posits that market forces
tend to increase, rather than decrease regional inequalities (1957). Once a location has
gained an initial advantage, economic growth tends to become concentrated there due to
advantages such as economies of scale and greater specialization. These places become
core areas, centers of economic activity like the manufacturing belt in the northeastern

United States or the Mexico City region.

94

Concomitantly, the process of economic growth tends to produce two contrasting
outcomes — spread effects and backwash effects. Spread effects represent positive
outcomes such as the economic development of peripheral locations due to the continued
expansion of core areas. Backwash effects refer to processes that bring about growth of
core regions at the expense of the further decline and marginalization of peripheral areas.
Whereas spread effects are most likely in economies that have achieved a certain level of
economic development, backwash effects may be more likely, at least initially, in
developing countries (Chorley and Haggett, 1967).

The widespread growth of complementarity throughout all regions of the United
States and the substantial improvement in transferability among geographically and
economically peripheral states may provide evidence of the spread effects of NAFTA.
Conversely, the increasing concentration of demand among states in northern Mexico, in
conjunction with the declining complementarity of states in the southern part of the
country, may be an indication of Myrdal’s backwash effects.

The “new” trade theory, associated primarily with Krugrnan (1994, 1995), is
based on Myrdal’s notion of circular and cumulative causation — increasing returns,
economies of scale, and greater levels of specialization. The recent research of Hanson
(1996, 1997), in particular, provides an indication of how trade liberalization measures
such as NAFTA may have impacted upon the economic geography of the United States
and Mexico.

Hanson (1997) asserts that trade policy plays a critical role in regional economic
development and may diminish regional disparities in industrial employment and wages.

However, he cautions that an even pattern of development will not result because trade

95

agreements may shift industry to regions closer to foreign markets. In the case of Mexico,
Hanson concludes that trade policies have prompted the decentralization of industrial
activity ﬁ'om the area in and around Mexico City to the northern part of the country,
especially along the US-Mexico border.

Analysis of changes in complementarity between 1994 and 1997 clearly supports
Hanson’s conclusions. From a geographic perspective, the locational changes in the
demand of US imports among states in the northern part of Mexico is the likely result of
the decentralization of economic activity following implementation of NAFTA in 1994.

Hanson (1996) also asserts that trade policies have had a signiﬁcant impact on the
economic geography of the United States. In fact, his research contradicts the view that
changes in the Mexican economy are too small to have any real impact on the US
economy. As industrialization has expanded in Mexican border states (the maquiladora
program, primarily), manufacturing activity also has expanded in adjacent US border
states. Notwithstanding the ongoing restructrning of the US economy, Hanson predicts
that manufacturing activity will continue to relocate from the traditional manufacturing
belt to locations closer to the border.

As this thesis reveals, the increasing relative complementarity of states in
southern and western parts of the United States is likely evidence of the ongoing
restructuring of the US economy. However, as Hanson’s research suggests, locational
changes between 1994 and 1997 in commodity groups such as chemicals, textiles, metal
products and machinery (largely intermediate goods) also may be the result of decisions

within speciﬁc industries to take advantage of industrial expansion in northern Mexico.

96

In conclusion, although this thesis illustrates problems in the calibration of gravity
models with incomplete interaction matrices, it also indicates the potential promise of
examining changes in the geographic, as well as commodity, structure of trade. Though
the relative accessibility and importance of centers of supply and demand in both
countries has not changed markedly, this thesis indicates that spatial interaction between
the two countries has grown signiﬁcantly since inception of NAFTA. These changes in
the geographic structure of trade have generally been overlooked in the literature to date.

The results of this thesis also provide empirical support of the “new” trade theory.
Although international economists have typically focused on changes in regional
employment and wage structures to provide evidence of increasing returns and greater
specialization, this study presents a geographic basis, focusing on notions of
complementarity and transferability, in order to assess locational changes in economic
activity.

This thesis also reveals several potential areas for further research. The limited
explanatory power of traditional gravity models indicates the need to develop spatial
interaction models that permit more comprehensive assessment of changes in the
accessibility of origins and destinations and identify factors other than supply, demand
and distance that detennine levels of trade between US and Mexican states. In addition,
as Hanson’s research suggests, an opportunity exists for economic geographers to identify
the spatial consequences of national policy decisions (such as NAFTA) and “new” forms

of international trade on regional economies.

97

BIBLIOGRAPHY

Bibliography

Abler, R., Adams, J ., and Gould, P. 1971. Spatial Organization: The Geographer's
View of the World. Englewood Cliffs, NJ: Prentice-Hall.

Alonso, W. 1980. A Theory of Movements. in Hansen, N., ed. Human Settlement
Systems: International Perspectives on Structure, Change and Public Policy.
Cambridge: Ballinger.

Batten, D. and Nijkamp, P.1990. Barriers to Communication and Spatial Interaction.
Annals of Regional Science, vol. 24, pp. 233-236.

Baxter, RS. 1976. Computer and Statistical Techniques for Planners. London:
Methuen.

Black, W.R. 1971. The Utility of the Gravity Model and Estimates of its Parameters in
Commodity Flow Studies. Proceedings of the Association of American
Geographers, vol. 3, pp. 28-32.

Black, W.R. 1972. Interregional Commodity Flows: Some Experiments with the Gravity
Model. Journal of Regional Science, vol. 12, no. 1, pp. 107-118.

Black, W.R. 1974. An Analysis of Gravity Model Distance Exponents, Transportation,
vol. 2, pp. 299-312.

Boyer, K.D. 1996. American Trucking, NAFTA and the Cost of Distance. Annals,
American Academy of Political and Social Sciences, vol. 553, pp. 55-65.

Byler, J. and O'Sullivan, P. 1974. The Forecasting Ability and Temporal Stability of the
Coeﬂicients of Gravity Models Applied to Truck T raﬂic. Trafﬁc Engineering
and Control, pp. 474-476.

Chavez, M. and Whiteford, S. 1996. Beyond the Market: Political and Socioeconomic
Dimensions of NAFTA for Mexico, in Roberts, K. and Wilson, M., Policy
Choices: Free Trade among NAFI‘A Nations. East Lansing: Michigan State
University Press.

98

Chishohn, M. and O'Sullivan, P. 1973. Freight Flows and Spatial Aspects of the
British Economy. Cambridge: Cambridge University Press.

Chorley, R]. and Haggett, P. 1967. Socio—economic Models in Geography. London:
Methuen and Co., Ltd.

De Janvry, A., Sadoulet, E. and Davis, B. 1997. NAFTA and Agriculture: An Early
Assessment. Working Paper No. 807. Berkeley: Giannini Foundation for
Agricultural Economics.

Desta, E. 1988. Spatial Structure and Spatial Interaction. Ph.D. Dissertation.
Department of Geography. Michigan State University.

Desta, E. and Pigozzi, B.W. 1991.Further Experiments with Spatial Structure Measures
in Gravity Models. Tijdschrift voor Economische en Sociale Geografie, vol. 82,
no. 3, pp. 220-226

Dicken, P. and Lloyd, P. 1990. Location in Space: Theoretical Perspectives in
Economic Geography. New York: HarperCollins.

Feenstra, RC. and Hanson, G.H. 1997. Foreign Direct Investment and Relative Wages:
Evidence from Mexico ’3 Maquiladoras. Journal of International Economics,
vol. 42, pp. 371-393.

Fotheringham, AS. and O'Kelly, ME. 1989. Spatial Interaction Models: Applications
and Formulations. Dordrecht: Kluwer Academic Press.

Gordon, LR. 1985. Economic Explanations of Spatial Variation in Distance Deterrence.
Environment and Planning A, vol. 17, pp. 59-72.

Haggard, S. 1995. Developing Nations and the Politics of Global Integration.
Washington, DC: Brookings Institution.

Haggett, P., Cliff, A., and Frey, A.l977. Locational Analysis in Geography. New York:
John Wiley and Sons.

Haggett, Peter. 1983. Geography: A Modern Synthesis. New York: Harper and Row

Hanink, D. 1994. The International Economy: A Geographical Perspective. New
York: John Wiley and Sons.

Hanson, G.H. 1996. Economic Integration, Intraindustry Trade and Frontier Regions.
European Economic Review, vol. 40, pp. 941-949.

99

Hanson, G.H. 1996b. Localization Economies, Vertical Organization and Trade.
American Economic Review, vol. 86, no. 5, pp. 1266-1278.

Hanson, G.H. 1997. Increasing Returns, Trade and the Regional Structure of Wages.
Economic Journal, vol. 107, pp. 113-133.

Harris, C.D,.l954. The Market as a Factor in the Localization of Industry in the United
Sates. Annals of the American Association of Geographers, vol. 44, pp. 315-
348.

Haynes, K. and Fotheringham, AS. 1984. Gravity and Spatial Interaction Models.
Beverly Hills: Sage.

Helliwell, J.F. 1996. Do National Border's Matter for Quebec's Trade? Canadian
Journal of Economics, vol. 29, no. 3, pp. 507-521

Hinojosa-Ojeda, R. 1996. North American Integration Three Years after the
NAFTA: A Framework for Tracking, Modeling, and Internet Accessing the
National and Regional Labor Market Impacts. University of California at Los
Angeles. North American Integration and Development Center.

Hirschman, AG. 1958. The Strategy of Economic Development. New Haven, CT: Yale
University Press.

Hufbauer, GO and Schott, J .J . 1992. North American Free Trade: Issues and
Recommendations. Washington, DC: Institute of International Economics.

Isard, W. 1956. Location and Space Economy. Cambridge: MIT Press.

Isard, W. 1960. Methods of Regional Analysis: An Introduction to Regional Science.
Cambridge: MIT Press.

Keeble, D., Owens, PL, and Thompson, C. 1982. Regional Accessibility and Economic
Potential in the European Community, Regional Studies, vol. 16, no. 6, pp. 419
432.

Knudsen, D. and F otheringharn, AS. 1986. Matrix Comparison, Goodness of F it and
Spatial Interaction Modeling. International Regional Science Review, vol. 10,

no. 2, pp. 127-147.

Knudsen, D. 1988. On the Stability of Trade Partnerships. Environment and Planning
A, vol. 20, pp. 1335-1343.

Krugman, Paul. 1994. Rethinking International Trade. Cambridge: MIT Press.

100

Krugman, P. 1995. Development, Geography and Economic Theory. Cambridge: MIT
Press.

Krugman, P. 1993. The Uncomfortable Truth about NAFTA. Foreign Affairs, vol. 72,
pp. 13-19.

Linnemann, H. 1966. An Econometric Study of International Trade Flows.
Amsterdam: North Holland Publishing.

Losch, A. 1954. The Economics of Location. New York: John Wiley and Sons.

Lowe, J.C. and Moryadas, S. 1975. The Geography of Movement. Prospect Heights, IL:
Waveland Press, Inc.

Lustig, N., Bosworth, B., and Lawrence R. 1992. Assessing the Impact of North
American Free Trade. Washington, DC: Brookings Institution.

MacKay, R. 1958. The Interactance Hypothesis and Boundaries in Canada: A
Preliminary Study. The Canadian Geographer, vol. 2, pp. 1-8.

McCallum, J. 1995. National Borders Matter: Canada- US Regional Trade Patterns, The
American Economic Review, vol. 85, no. 3, pp. 615-623.

Mydal, Gunnar. 1957. Rich Lands and Poor: the Road to World Prosperity. New York:
Harper and Row.

Nijkamp, P. 1986. Handbook of Regional and Urban Economics. Amsterdam: North
Holland.

Nozick, Linda K. 1996. Trade Between the United States and Mexico. Transportation
Quarterly, vol.50, no.2, pp.l 11-133.

O'Sullivan, P. and Ralston, B. 1974. Forecasting Intercity Commodity Transport in the
USA. Regional Studies, vol.8, pp.191-195.

Pastor, RA. 1993. Integration with Mexico: Options for US Policy. New York: The
Twentieth Century Fund Press.

Rietveld, P.1993. International Transportation and Communication Networks in Europe,
The Role of Barrier Effects. Transportation Planning and Technology, vol. 17,
pp. 3 1 1-317.

Senior, M. 1979. From Gravity Modelling to Entropy Maximizing: A Pedagogic Guide.
Progress in Human Geography, vol. 3, no. 2, pp. 179-210.

101

Sheppard, E. 1984. The Distance-Decay Gravity Model Debate, in Gaile, G. and
Willrnott, C.J. Spatial Statistics and Models. Dordrecht. Reidel Publishing Co.

Studenmund, AS. 1992. Using Econometrics: A Practical Guide. New York:
HarperCollins.

Taaffe. E.J., Gauthier, BL, and O'Kelly, ME. 1996. Geography of Transportation.
Upper Saddle River, NJ: Prentice-Hall.

Ulhnan, EL. 1980. Geography as Spatial Interaction. Seattle: University of
Washington Press.

Warntz, W. 1959. Toward a Geography of Price. Philadelphia: University of
Pennsylvania Press.

Wilson, AG. 1970. Entropy in Urban and Regional Systems. London: Pion, Ltd.

Yeates, M. 1969. A Note Concerning the Development of a Geographic Model of
International Trade. Geographical Analysis, vol. 1, pp. 399-403.

102

"IIIIIIIIIIIIIIIIIIII