The Web Book of Regional Science

Sponsored by

Regional Development:
Challenges, Methods, and Models
By

Randall W. Jackson, West Virginia University

Geoffrey J.D. Hewings, University of Illinois
Serge Rey, University of California-Riverside
Nancy Lozano-Gracia, World Bank

Published: 2019

Editor: Randall Jackson

Director, Regional Research Institute

4
The Web Book of Regional Science is o ered as a service to the regional research community in an e ort
to make a wide range of reference and instructional materials freely available online. Approximately 30
books and monographs have been published as Web Books of Regional Science. These texts covering
diverse subjects such as regional networks, land use, migration, and regional specialization, include
descriptions of many of the basic concepts, analytical tools, and policy issues important to regional
science. The Web Book was launched in 1999 by Scott Loveridge with Regional Research Institute
directors serving as Web Book editors. Scott Loveridge performed that role through 2000 and Randall
Jackson served as editor from 2001 through 2022.
Web Book material, including text and graphics, is available to users for personal use and may not be
redistributed in whole or in part, in print, online, or on electronic media (e.g., CD). Permission for
reprinting images and text from the Web Book of Regional Science must be obtained from the Regional
Research Institute to which all users must comply.

When citing this book, please include the following:

Jackson, R.W., Hewings, G.J.D., Rey, S., & Lozano-Gracia, N. (2019).

5
The recognition that aspatial macroeconomic perspectives can prove to be misleading is far from new,
but there has been a recent and welcome increase in attention to the spatial dimensions of economic
development. Beveridge (1944) was among the rst to introduce a spatial dimension by examining the
role of full employment in a society with a special focus on the spatial distribution of employment and
unemployment. Readers were startled to nd that during the Great Depression (1929-1937),
unemployment rates varied by a factor of two to two and a half times between the Southeast of the UK
(London-centered region) and Wales, Scotland, and the North of England. The impact of the Great
Depression turned out to have been very uneven over space; the Beveridge book sought to understand
the causes and the outcomes. Eighty years later, during a period in which a variety of initiatives had been
undertaken, the regional disparities have persisted notwithstanding the introduction of signi cant
initiatives by the national government to address these disparities. For the three months ending in July
2017, the highest unemployment rate was still in the Northeast (6.0%) with the lowest recorded in the
Southeast of the UK (3.2%). 1 Many other places around the World have experienced similar deep
regional inequalities, with disparities between rural and urban areas playing a central role in such
inequalities (World Bank, 2014; Deuskar et al., 2015). In China for example, the rural-urban divide is
estimated to account for 45 percent of overall inequality (Kanbur et al., 2014).
Much of the disappointment with the outcomes from spatial interventions may be traced to a lack of
understanding of how regional economies work. In retrospect, it would appear that many of the
initiatives aimed at reducing regional disparities have been proposed with an incomplete understanding
of how they might work; in many cases, there was little or no attempt to measure the e cacy of the
initiatives. In other cases, the lack of consideration of spatial/regional disparities may have compromised
the e cacy of broader development policies. Too often, there is a na ve assumption that investment in a
region can only generate bene ts or costs in that region; spatial spillovers are, more often than not,
ignored. A distinguishing feature of regional economies is their openness, which leads to complex webs
of interregional linkages that tie the fortunes of individual regional economies together. This means that
one region cannot be studied in isolation from the broader interregional system, and that measurements
of interregional linkages and the impacts of these linkages on regional economies are required.
There have been signi cant changes in the structure of national and regional economies in recent
decades that have challenged some of the assumptions on which many regional policy interventions
were once proposed. All the while, regional disparities in levels of welfare, growth rates, employment
and unemployment have persisted, and one of the major challenges in most countries centers on
addressing this persistence. However, earlier views that there is a negative trade-o between national e
ciency and regional equity remains current, leading many policy analysts to view attention to regional
problems as a cost to overall national growth and development.
The spatial dimensions of economic activity have received increasing attention in the last 20 years (Fujita
et al., 2001). Parallel to these developments in economics has been the widespread adoption of
geospatial analysis and technologies across the wider social sciences (Goodchild et al., 2000), which re
ects the ability of these methods to provide insights as to the role of spatial processes and structure in
the operation of many socioeconomic phenomena, and in turn, more comprehensive understanding of
the world.
In this paper we review the challenges that the consideration of regions brings into economic analysis
and provide an overview of some of the key methods and tools that can be used to gain a better

1 [Link]
bulletins/regionallabourmarket/september2017. Accessed 12/04/2018.

1
understanding of how regional economies work, and through that, identify both the challenges and
opportunities that they face. In the next section, the exploration of these challenges begins with some
consideration of the ways in which regional economies work to set the stage for subsequent sections
that summarize a toolbox of methods and strategies that might be considered for both ex ante and ex
post evaluation of regional development initiatives. In contrast to past reviews of this eld, this report
presents an integration of more traditional regional macroeconomic modeling with new developments
in spatial data analysis.

1 How Regional Economies Work

Brakman et al. (2010) suggest that regional economics ... analyzes the spatial organization of economic
systems and must somehow account for the uneven distribution (of economic activities) across space...
In contrast, regional growth and development theory attempts to answer three main questions:
Why and how do regions grow?
Why do some regions grow more rapidly than others?
Why do some regions decline?
Unevenness in the neoclassical view of Borts and Stein (1964) would not be a problem since mobile
factors (such as labor and capital) would move to extract their maximum rents, leading to a world in
which an equilibrium distribution would see no di erences among regions. For example, workers would
move to regions with higher wages causing labor market adjustments in origin and destination regions
until wages equalize across regions.
Many of the neoclassical regional models were based on a Ricardian notion of comparative advantage, in
which a region exports those goods that it produces at relatively lower production costs that result from
di ering relative productivities of the factors used to manufacture the goods. It is trade in goods or
factors, adjusting through their relative prices and increased productive capacity, that results in the
achievement of full employment. In this world, di ering remunerations of the production factors
reallocate resources in space, and thus generate a higher rate of growth and eventually, in equilibrium,
equalized factor returns.
However, factors are neither in nitely divisible nor homogenous and certainly not perfectly mobile. This
leads to non-equilibrium outcomes in which di erences in levels of welfare between regions can arise
and persist. So, what happens when mobility is limited, and instead of constant, there are increasing
returns to scale, with di ering initial endowments and number of sectors in each region?
According to Richardson (1978), regional growth economics di ers from aggregate growth theory by its
emphasis on interregional factor mobility as a key growth determinant; neoclassical regional growth
models simultaneously o er explanations of endogenous system of growth and interregional factor ows
with a single model. However, as regional economics developed, it became clear that many of the
assumptions in neoclassical theory were di cult to envisage operating in reality; in fact, Armstrong et al.
(2000) advance the notion that one of the reasons that regional development policy can be justi ed is to
address market failures.
Among the many critical assumptions of traditional neoclassical models, is the one that a common wage
level prevails in ALL regions because of perfect adjustment on the demand side due to di erentials in
labor supply growth. However, frictions to the mobility of labor are common in many parts of the world.

2
Further, in some regions, labor supply grows faster because of a faster rate of natural increase and/or
net migration, and this faster increase in labor supply or barriers to mobility of factor markets can have a
dampening e ect on wage rates and can prevent equalization of wages.
Under the assumptions of the neoclassical model, even if a region produces all goods at higher prices, so
that it is generally more ine cient in its production processes than any other region in the country, it may
nevertheless be relatively less ine cient in producing one particular good. The region will thus be able to
obtain a role for itself in the international division of labor by specializing in the production of the good
in which it is relatively more e cient. As Capello (2015) has noted, this argument has major normative
implications, for it asserts that there is always an automatic mechanism guaranteeing the existence of
some specialization, regardless of productive e ciency, and therefore economic policy measures to foster
development are unnecessary.
However, it is the imbalance in interregional factor endowments, and di erences in levels of factor
productivity, that account for the advantage enjoyed by a local system in its relations with the rest of the
world. These are the elements that underlie a region’s growth path and condition its timing and the form
that it takes. Capello (2015) notes that there are numerous sources of territorial competitiveness: one of
the main foci has been on role of di erences in factor endowment as the main source of territorial
competitiveness.
More generally, many of the limitations of the neoclassical model and its reliance on automatic
equilibrating mechanisms stem from two assumptions: that (1) the capital market is perfect; and (2)
technical progress is available to all regions no one region has an advantage. Under these assumptions,
the outcome is convergence in regional incomes. However, empirical evidence refutes the conclusion
that capital ows to regions with low income levels. And instead, most countries end up divided into
leading regions where investment and economic activity concentrate, and lagging areas where
investment remains low, economic activity is incipient, and basic living standards are lacking.
An alternative approach that would furnish a better interpretation of the real movement of the factors
would con rm the tendency of capital to shift to areas with higher wage remunerations; in this
alternative approach, often referred to as the Two-Sector model, subsequent reallocation of resources,
due to an external shock that moves the regions far from the initial steady-state equilibrium, pushes
local economies towards permanently di erent growth rates.2

Consider a two-region case and assume an initial equilibrium, with the growth rate stable and uniform
between the regions and in which capital and labor grow in each of them at a constant rate equal to that
of income. Now consider an exogenous shock, for example, wherein the demand for the good exported
by one of the two regions increases. The price of the good rises as a direct consequence and this e ect
has a positive impact on the value of the marginal product of the factors in the region. The outcome is an
intra- and interregional reallocation of production resources. The amount of capital stock increases in
the producing sector to meet the increased export demand because of the in ow of external capital
attracted by greater returns. However, labor demand by local rms will also increase because of the
increase in the value of the marginal product of labor (generated by the rise in the exported good price).
There will be two further e ects, one generated by in-migration and the other by labor switching. The
greater demand for labor will attract workers both from the local agricultural sector and from other
regions, given the higher remunerations that will be available. Expansion of production and employment
in the sector producing for export will thus have a backwash e ect on the agricultural sector. The main

2 Recall that the neoclassical model has production factors migrating because they are attracted by higher remunerations.

3
outcome is regional divergence the persistently higher growth rate in the region for which export
demand is increasing generates attractive forces for both labor and capital that make it di cult to envision
a situation in which the agricultural sector will catch up in terms of income convergence, without
additional interventions. While the less prosperous regions o er advantages due to their relatively lower
wages and unit labor costs, these are often not su ciently attractive for signi cant relocation of economic
activity.
Persistence of marked regional disequilibria suggests that these locational advantages are not enough to
close the gap between advanced and backward regions. The more prosperous areas are able to absorb
the decreasing returns that accompany industrialization and high capital intensity, while the less
prosperous regions of the advanced countries (such as those regions in the new accession countries of
the EU) have to compete with the low unit labor costs characteristic of the developing countries. These
issues highlight an emerging problem in regional analysis the existence of di erences at di erent spatial
scales (e.g., EU versus the developing world at one level, di erences between regions within the EU and
even important heterogeneity within regions, for example between urban and rural areas, and even
within neighborhoods in a city).
Further, within a country, the economic performance of a region depends largely on its absolute rather
than its comparative advantage. A negative shock in a national sector is often met with real depreciation,
which continues until other sectors become competitive. Within a country, labor markets are often more
integrated than across countries, and hence if a region su ers a negative shock, exibility of adjustment of
relative wages will be lower making equalization of wages harder. Prices of immobile factors like land
and housing may decline but this may not be enough to bring the local economy to competitive levels,
suggesting persistent regional disparities and possible loss of labor rather than the convergence
mechanism often expected at the national level (Duranton and Venables, 2018).
Another issue not addressed by the neoclassical model is the di culties of factor ow movements, since
the model assumes that the costs of migration are zero, when migration costs can often be high. This will
be true not only of labor, but also for rms - especially those whose production systems require signi cant
investment in buildings and associated infrastructure. In addition, full information may not be available
and even when available, Basile and Lim (2017) have identi ed a non-linear relationship between
migration and wage di erentials, with an inertia range in which little migration is observed until the di
erentials become large enough that people do move. One might appeal to the role of assets or access to
credit as an important part of the decision-making process in making a move. Capital, even if mobile,
may remain in more prosperous regions because of cumulative processes, information spillovers relating
to innovation and other factors that the New Economic Geography (NEG) groups into agglomeration
forces. Further, labor may not be able to move from agriculture to industry because of lack of skills
generating the possibility of wages increasing in one sector even with high unemployment in the other.
While the two-sector model has still some limiting assumptions, it does move the discussion closer to
empirical realities and toward a better understanding of the speci cities of regional economics.
Regional analysis has reached a stage in which theory has formed the foundation for a better
understanding of how spatial development happens, incorporating some of the frictions that we see in
reality and that neoclassical theory ignores. Yet, regional economic development theory still falls short of
complete spatial-awareness of explaining both why and where things are happening.
Policy proposals for regional development have followed a similar process of evolution, sometimes
developing hand in hand with theory and sometimes despite it. In the next section, a brief review of

4
some past and current policies will be provided, all premised on the notion that market failure, in the
form of uneven development, requires some form of intervention.

1.1 An overview of policy proposals

In the 80 years since the publication of Beveridge’s monograph there has been a variety of regional
policy proposals aimed at the reduction of interregional disparities. Many of these have focused on
limited development in more prosperous areas while others focused on enhancing the competitive
potential of the least prosperous regions of a country. An excellent but somewhat dated overview may
be found in Armstrong et al. (2000) with some of the more recent policies reviewed by McCann (2015),
with a special focus on the United Kingdom and the European Union.

The Early Regional Policies-Sticks and Carrots

In the 1960s, especially in the United Kingdom, regional policies oscillated between those termed
Carrot and those characterized as Stick. The former provided incentives for rms to locate establishments
in less prosperous regions; these incentives ranged from outright grants to tax abatements and funds for
the training/retraining of employees. In analysis of the Department of Regional Economic Expansion
policies in Canada, Woodward (1973, 1974, 1975) characterized these policies as windfalls; in many
cases, the rms had already decided to locate in less prosperous areas (especially in the Atlantic
provinces) and so the incentives provided little bene t to these economies but were clearly attractive to
the rms (see a more recent comprehensive evaluation by Scho eld (1989) and PolŁse and Shearmur
(2006a,b). In contrast, the stick policies were focused on restricting development in the most prosperous
parts of the country (e.g., London in the United Kingdom). The argument advanced was that without
such restrictions, rms would continue to place establishments in these regions and thus exacerbate the
already profound existing disparities. Neither policy was particularly e ective, and few countries adopted
the restrictive policies, but many continue to o er subsidies and incentives for rms to locate facilities in
their region (referred to globally as scal wars such as the one recently visible in the US for the location of
Amazon’s second major headquarter facility). Rarely, does one nd careful project appraisal techniques
being used to evaluate the expected return on investment by public agencies although US states are
becoming more assertive in tying incentives to a set of demonstrable metrics (e.g., numbers of jobs
created).

Growth Poles and Growth Centers

The fascination with Perroux (1955) concept of p les de croissance (growth poles) generated a new wave
of interest in the creation of innovative policies that address regional disparities. Since Perroux’s concept
was aspatial, proposing to exploit the dynamics of critical linkages in an economic system, it was quickly
translated to the term growth centers. However, like Perroux’s original concept, de nition proved elusive
to the point that growth centers were often designated ex post. What was retained was a sense of the
role and importance of sectoral linkages, building on some of the early ideas of the notion of keys
sectors (Hirschman-Rasmussen). One of the main problems of this set of ideas is that they looked to the
past rather than providing indicators about the future growth and development potential of a region (on
this see Diamond, 1974).
Portfolio Theory

5
For a number of years, regional development strategy became enamored with an industrial policy
analogous to nancial portfolio theory in which the explicit trade-o between risk unemployment volatility
in the industrial case and return employment growth by industry was applied to regional economic
systems. The region’s industrial portfolio was to be viewed as though it were a stock portfolio, which was
then evaluated in terms of two dimensions the expectation for growth and the risk associated with
public capital investment in the sense that a faster growing sector might also be one that experiences
signi cant cyclical behavior (e.g., boom followed by bust). The mathematical programming portfolio
variance techniques already in application in nance could then be applied to seek a mix of industries that
would provide an acceptable balance between risk (economic uctuations) and return (growth rates).
Policy makers could indicate preference for risk minimization by supporting the development of
industries whose employment uctuations were countercyclical and o setting, or for more rapid growth
with less concern for employment or unemployment volatility. Following its introduction by Conroy
(1974), the approach attracted support from early proponents such as St. Louis (1980) along with some
additional commentary and criticism o ered by Jackson (1984). The idea seemed to have experienced a
modern renaissance with work by Chandra (2003) and Chiang (2009) but has been eclipsed by the
cluster-based strategy most closely associated with the work of Porter (1990).

Industrial Complexes and Cluster Strategies

In regional science, there had been an emerging tradition in industrial complex analysis in which
attention was directed to the interconnections between industries within some a priori de ned
geographic space (see Czamanski, 1971; Czamanski et al., 1974; Czamanski and Czamanski, 1976 and
Czamanski and Ablas, 1979). The in uence of Marshallian ideas of industrial districts and the externalities
they generate was also in uential but, as Martin and Sunley (2003) have noted, Porter was perhaps not
as generous as one might have wished him to be in acknowledging the intellectual foundations of his
proposals. Porter’s initial ideas centered on a diamond of interactions within a local context rm strategy
and rivalry, factor inputs, demand conditions and related and supporting industries. The latter three
components could be found in Czamanski’s work, for example, while Porter’s addition of rm strategy and
rivalry clearly di erentiated his work from traditional academic endeavors and positioned him to
command the attention of national and regional governments eager to embrace a new strategy for
enhancing competitiveness. Martin and Sunley (2003) also re ect a growing sense concern about a
concept that has not been rigorously evaluated. While this is a valid concern, it applies equally to
virtually all development strategies. While the relationships between structure and performance can be
assessed, on this basis alone one cannot really conclude anything about whether there was an explicit
development program that had consequences consistent with the observed relationships or changes. In
fact, extensive searches of the literature fail to produce evidence of rigorous evaluations for example,
the adoption of some quasi-experimental design in which two similar regions, one with and one without
a clusterbased development strategy, could be compared in terms of some appropriate metrics of
economic performance.
This is not to suggest, however, that there is evidence suggesting the failure of cluster development
strategies. Indeed, cluster strategies themselves are often not formally de ned (Yu and Jackson, 2011),
and can include any combination of state-sponsored e orts, groups composed of private entrepreneurs
working to enhance interindustry linkages, actively promoted cluster initiatives, and clusters that have
developed essentially organically, with the most prominent example of the latter kind being Silicon
Valley. Woodward (2012) describes the success of the South Carolina auto cluster, which continues to
gain momentum, and there is anecdotal evidence that many other cluster initiatives have been

6
successful. Cluster initiatives vary from case to case, of course, complicating their evaluation and
assessment. The most substantial contributions from some cluster initiatives can simply be the
establishment of and focus for networks of like-minded individuals working toward a common goal.
Cluster-based initiatives that engage more and more diverse actors in their networks, such as academic
institutions, chambers of commerce, governmental bodies, and private entrepreneurs would be
expected to be more e ective than those that engage fewer, but again, empirical assessment is lacking.

The Role of the Creative Class: A New Take on Key Sectors

Similar criticisms surrounding a lack of evidence-based support might be directed towards Florida’s
(2005a) exposition of the role of the creative class; in his book, he proposes that while ... the role of
creativity in city formation and growth is new, ... with the decline of physical constraints on cities and
communities in recent decades, creativity has become the principal driving force in the growth and
development of cities, regions, and nations. As with many concepts that focus on a subset of a regional
or urban economy, there is always going to be criticism directed to ignoring or downplaying the role of
the rest of the economy. As Diamond (1974) once commented in connection with the concept of key
sector analysis (of which the creative class idea shares many conceptual similarities), a case may be
made that non-key sectors may be more critical since, without them, the key sectors would not be so
prominent. A more nuanced vision of an economy might appeal to the one proposed much earlier by
Leontief who highlighted the roles of circulation and interdependence and drew attention to the notion
of complexity generated by consideration of not only direct relationships but the importance of indirect
ones (something not considered in Porter’s work).

Smart Specialization
More recently, there has been interest in capturing some of the characteristics of earlier policies within a
more comprehensive framework that is referred to as smart specialization (see McCann, 2015).
Conceptualized as a new way to integrate innovation and development policy, the proposed strategy was
motivated by concerns about the productivity gap between the European Union and the United States
that could be traced to the former’s weakness in the di usion of new knowledge and technologies across
sectors. The causes postulated included (1) di erences in labor markets; (2) management performance;
(3) organizational issues and (4) market deregulation. According to McCann (2015), the key turned out to
be the critical role played by new information and communication technologies (ICT)since it was ICT-
producing sectors that were driving the productivity gap. Smart specialization was originally a non-
spatial construct focused on the idea of a knowledge ecology the endogenous context in which
technological evolution takes place, focusing on pathways for innovation, the evolutionary experience of
the system, inherited structures, institutions and actors and their interrelations. In addition, it focused
on the role of entrepreneurial search processes and the domain in which innovation takes place
especially its connectedness, thus highlighting the role of R&D in the broadest sense.
Further, smart specialization highlighted the role of domain, relevant size (scale economies) and level of
connectedness and in this sense drew on ideas from NEG but with much greater focus on the
relatedness of varieties. It appealed to the attraction of Marshallian externalities, namely that it would
be easier to switch between related than between unrelated activities (in response to changes in
demand). All these attributes needed to be embedded in the domain with a focus on local linkages,
dependence on local skills or local institutions.

7
Connectivity highlights the degree to which rms are linked within the domain via transport, interpersonal
and nancial links. The policy implications stress the proposition that in order to foster innovation and
growth, regions should prioritize those activities enhancing entrepreneurial search activities to diversify
those that are highly embedded in the region. There should be a concerted attempt to build on existing
dominant technological and skills pro les and capabilities but then diversify around this core base. While
this proposal might be appropriate for those regions that actually have dominant tech and skills pro les
and capabilities, it would be di cult to translate into action in places that lack them such as NE Brazil,
perhaps, or even a state such as West Virginia in the US. Empirical evidence suggests that regions grow
more successfully and are more resilient if their growth patterns are more related to their historical
trajectories. Hence, in contrast to portfolio theory, the diversi cation strategy is more nuanced in that it
does not posit a sharp break with prior specialization. Smart specialization does not imply regions should
become more specialized because this will make regions more vulnerable to shocks. In essence, regional
policies should promote not just diversi cation but specialized diversi cation. Particularly astute are the
obvervations of Malizia and Feser (1999), who note that, The economic diversity of a city can be de ned
in reference to its specializations. As additional relatively independent specializations co-locate, the area
becomes more diverse. Economic diversity is the presence of multiple specializations. In response,
Jackson (2015) developed a method designed to identify new activities that move the regional economy
toward specialized diversi cation. This approach contrasts with some earlier attempts to explore diversi
cation to provide some pro-active counter-cyclical resilience in regions that were characterized by a
relatively narrow economic base. For example, in an evaluation of a diversi cation policy for Appalachia,
there was no consistent superior performance in counties that were more diversi ed than others. In
many cases, the more diversi ed counties had become that way as a result of the loss of specialized
industrial activity rather than through the addition of rms in other sectors. Indeed, Chinitz (1961) urged
the careful assessment of how changes in one industry might impact regional suitability for others.
Chinitz fell just short of explicitly identifying a paradox for the structure to performance relationship.
Namely, should an anchor industry in a region su er a sharp decline, the region by most measures would
as a direct consequence become more diversi ed, though certainly not better o (Jackson, 2015, p. 119).
Whatever polices that are proposed, there is a concomitant need to provide an analytical framework
that can be used for evaluation. Evaluation is signi cantly underrepresented in the literature in large part
because of the di culty of demonstrating that policy x has made a di erence. Singleequation analysis
(e.g., Moore and Rhodes, 1974) is clearly not complete and the di culties of comparative analysis remain.
Further, as regions become more interconnected, comparison based only on internal structure fails to
highlight potential di erences in external linkages. In subsequent sections, a set of available models and
approaches that could be used for policy analysis will be identi ed.
Academics and policy makers are reluctant to highlight failures and as a result, many policies continue to
be recycled despite their often-limited e cacy. Additional work is needed to assess what has been
learned from previous policies about what did and did not work. What made a di erence and in what
ways? How did the outcome vary by the type of intervention, such as a direct impact (e.g. a grant or
subsidy) or an indirect one (infrastructure, investment in human capital etc.)? What were the positive
and what were the negative impacts of demographic changes (aging, composition of the labor force,
migration, occupational/skill capital) and economic development? An example of such assessment is
done in recent work by Rodr guez-Pose and Wilkie (2018) where a series of policies to promote growth
and development in underperforming, lagging areas is reviewed and evaluated through a critical
discussion of the impacts of such policies. However, much more empirical evidence is needed to build a

8
body of work that can suggest with con dence which policies have had the desired impacts and which
have not.

1.2 Avoiding one size ts all pitfalls in terms of diagnosis and policy prescriptions
One major theme that has emerged from the regional policy literature is that there are no universal
approaches that seem to work in all places. In large part, this problem can be traced to the idiosyncratic
nature in which national and regional economies interact, their history and traditions, the nature and
strength of government institutions and the degree to which notions of scal decentralization have been
adopted.
One common suggestion could be advanced, namely that before any policy is adopted, a careful
appraisal of the structure of the regional economies would be necessary. While this might seem self-
evident, there are many cases of policies being adopted prior to a diagnosis of the structure of the
regional economy. In the next section, a rational for this diagnosis will be presented; thereafter, the
discussion will move to a consideration of regional structure and some of the methods that can aid in
diagnosis and interpretation.

2 What Have Regional Policies Missed?

2.1 Critical nature of internal heterogeneity
Beveridge (1944) identi ed the nature of the internal heterogeneity of the United Kingdom economy, a
phenomenon that has persisted in almost all countries. The enormous expenditure of funds provided by
the European Union for cohesion policies provides further testimony to the critical nature and
persistence of inequalities in the level of welfare both across and within countries. As noted in the
previous section, it makes little sense to discuss the development of a country such as Brazil without
providing concomitant detail on the internal spatial distribution of economic activity and population. To
gain a deeper understanding and appreciation of heterogeneity within countries and across regions, it is
key to look at dimensions such as the trade of goods, services and people as that provides a sense of the
strength of linkages among interconnected regional economies. Innovations and income transfers will
also reveal critical aspects of interregional knowledge and resource linkages. Further, assessing
geographical, cultural, social, and infrastructure availability di erences can also shed light on where
disparities in living standards stem from and provide an avenue for breaching existing gaps.

2.2 Identifying spatial interconnections-trade in goods and services, trade in people,

innovations, interregional income transfers
Modern production systems have become increasingly more fragmented in recent decades, with di erent
phases in production systems often allocated to di erent geographical locations. As a result, interregional
and international trade ows have been growing at rates in excess of the corresponding rates of growth of
gross regional or national domestic product; further, this trade is increasingly dominated by intra-
industry rather than inter-industry trade, re ecting the exploitation of economies of scale in production
of speci ed components and the signi cant reduction in transportation costs that facilitate spreading
production across di erent locations along the value chain (Krugman, 1991). This process has been
propelled in part by a signi cant spatial reorganization of value chains over the past two or three
decades, and the concomitant logistical issues associated with the most e cient coordination of

9
production systems has generated a complex system of interdependent ows, linking regions in one
country with regions in another.
This process of hollowing out (namely, the substitution of external sources of inputs and sales for
intraregional transactions) has seen intra-economy multipliers decreasing while interregional spillovers
are increasing; this phenomenon is occurring at both the interregional and at the international scale.
Hence, one can no longer assume that a similar project would generate the same spatial and total
impact wherever it was located within the nation at hand. The development of multiregional input
output and computable general equilibrium models has revealed that, contrary to Friedman (2005), the
world inside nations is not at; space is spiky and it is uneven (see Florida, 2005b). Further, projects
generate di erent spatial distributive impacts depending on the nature (highway, new business,
investment in human capital) and on the location of the project. In addition, spillover effects are not
necessarily symmetric: a project in Cataluæa might generate larger impacts on Madrid than a project in
Madrid generates on Cataluæa. Major projects can disturb the spatial equilibrium as factors such as
capital and labor respond to changes in opportunities and rents by relocating.
As the processes of fragmentation and hollowing out continue, interregional dependency will assume
even greater importance in explaining the growth and development paths of economies. The tragic
events in 2011 in Fukushima, Japan (earthquake and tsunami) and in Thailand ( oods) revealed risks
associated with extensive supply chains that reach across many widely spread locations and showed that
disruptions in even the smallest components can generate severe stress on the whole productive system.
Regional economies are becoming both more competitive and more integrated/complementary at the
same time, creating new challenges for policy analysts. To understand these new challenges to economic
development, it is essential to develop and maintain tools such an interregional input output models
that can assist in tracking these changes; focusing on just one region can generate misleading outcomes.
As in almost all formal economic modeling, the distinction between endogenous and exogenous is very
important; in constructing single-region models, it is often assumed that the impacts of exogenous
change are of a top down nature with no feedback e ects. However, the changing structure of regional
economies has resulted in greater role for interregional trade and the possibility that feedbacks could
prove to be important.
In this context, attention to trade needs to be expanded from attention to goods and services to people,
ideas and information ows. In many cases, movement of people (e.g., out-migration) can often
exacerbate the challenges of a less prosperous economy as Vanderkamp (1971) noted many years ago.
The eld of spatial econometrics has developed an increasingly sophisticated set of tools to help identify
the nature and strength of interregional spillovers to complement some of the more traditional
interregional models that can now be constructed relatively easily.

2.3 Addressing socio-cultural barriers and increasing concerns about increased inequality
The publication of Capital in the Twenty-First Century refocused attention on the role of inequality within
nations (Piketty, 2017); but many prominent scholars, such as Stiglitz (2012) and Atkinson (2015) have
provided rigorous commentary on the impact of inequality. Piketty’s major contribution was to stress the
role of the di erential between returns to capital in contrast to labor as a major source of increasing
inequality. However, these discussions usually focus on interpersonal inequality, rarely mapping the
outcomes into a spatial context. There is no consensus about how inequalities should be reduced and
the measurement of the cost of inequality is still not generally accepted. Appeals to broader de nitions

10
of welfare that extend beyond monetary considerations are often used to justify some form of
intervention to address inequalities.

2.4 Challenging the regional equity-national e ciency trade-o solutions to some regional
problems may enhance national e ciency
Regional analysts have been exploring the nature (and even the existence) of a trade-o between
attention to regional problems and the impact this might have on national economic e ciency. Archibald
(1969, 1972) was one of the early contributors exploring this apparent trade-o , and subsequent work by
Thirlwall (1969, 1970), Mera (1967, 1973) and Thirsk (1973) found that the simple expectation of a
trade-o was often not con rmed in empirical work. In fact, Higgins (1973) was more forceful in suggesting
that:
Measures to reduce regional gaps, far from being a luxury to be a orded when things are
otherwise going well in the country, are the essence of a policy to accelerate growth, reduce
unemployment and maintain stability. For developing countries, where e orts to accelerate
growth are inhibited by fear of aggravating in ation, reduction of regional disparities may
well be the sine qua non of a successful development policy.
National and subnational policies necessarily di er for a variety of reasons. Prominent among these di
erences is that there are many fewer policy handles available to regional policymakers than there are to
national policymakers. Regional policymakers have no control over macroeconomic variables like money
supply or interest rates, they cannot regulate trade or migration, and they typically have smaller
amounts of and less discretion over funds that might be transferred from one to another sub-region.
Because regional o cials have much less control over cross-border transactions, they nd themselves in
the role of price taker rather than price maker. Regional economic systems are typically much more open
than national economies and depend on other national sub-regions for inputs to their production
processes, the provision of consumption and investment goods, and for markets for the goods and
services they produce.
As a result, policies at the regional level often focus more heavily on structural economic characteristics
on the one hand, and on social and economic infrastructure on the other. The healthiest and most vital
interregional economic systems are composed of healthy and vital regional economic systems. For this
reason, the starting point for most analyses of regional systems is focused on regional structure. Most
structural analyses address industrial structure, labor force characteristics, and household income and
demographic characteristics.

3 Exploring Regional Structure

Understanding the economic structure of a region can help understand where its development potential
lies, and where the key bottlenecks to growth might be. Better understanding of the economic structure
of a region can also help diagnose which links with other regions are more important, where
coordination is essential, and where important links might be broken and need attention. To
demonstrate some of the key characteristics of interregional systems that support taking a regional and
multiregional structural approach to national policymaking, consider the following idealized 3region
example. The three regions, I, II, and III are of decreasing size and socioeconomic conditions, as re ected
by per capita value added (see table 1).

11
 Table 1: Three Regions: Summary Characteristics
Population VA

Region (millions) (millions) PCVA

I 10 450 45.0
II 9 308 34.2
III 8 258 32.3
Nation 27 1016 37.6
By fully aggregating industry transactions, we can present the interregional accounting system in Table 2
where the interregional transactions can be seen in the upper left 3x3 quadrant, denoted Z, and in coe
cient form in matrix A.
Table 2: The Structure of Interdependence
I II III Y
IA = 320 63 48 56 0.3 0.09 0.08
II 9 2
III 40 21 54 39 0.0 0.30 0.09
VA = 0 6 4
Imports 50 28 16 35 0.0 0.04 0.28
X= 8 4 5
The diagonal values in Z are 450 30 25 aggregations of all interindustry transactions that
take place within the 8 8 regions, and the o -diagonal cells show between-
regions aggregations. 140 91 72 Value-added is presented here is a single VA row,
but is an aggregation of 100 70 60 employee compensation, payments to
government, and gross 0 0 0 operating surplus. Gross output, X, equals the
sum of payments by industries in the region to intermediate suppliers,
to value-added, and to the rest of the world for imports. The nancial units here can be thought of as
thousands of dollars for purposes of discussion, but the relationship between nancial units and
population could be scaled di erently to re ect actual national development levels without loss of
generality. This example was constructed in a way that would be consistent with nations that are
composed of larger, more developed and typically more urbanized regions, and successively smaller and
less developed regions. Such nations typically have a large urbanized area for which productivity is
higher and per capita incomes are likewise higher than in the two smaller, less urbanized or more rural
areas.
The interregional multiplier matrix (Table 3) for this national system is similarly representative.
Intraregional multipliers decrease from 1.5 in Region I, the largest and most developed region, to 1.45
and 1.39 in the smaller regions. Although the example was deliberately constructed to demonstrate the
potential for enhancing socioeconomic conditions in the lagging regions without sacri cing national
equity, the trade relationships shown are not unreasonable nor out of the ordinary. Region I is an
important supplier for all regions, and it purchases much more from itself than it does from the other
two regions. Likewise, Region II purchases more from dominant Region I than it does from Region III, and
Region III purchases almost equal amounts from both regions.
Table 3: The Interregional Multiplier Matrix

12
 I II III

I 1.50 0.20 0.19

II 0.10 1.45 0.19
III 0.11 0.09 1.39
Multiplier 1.70 1.75 1.77
We take the position here that the interindustry and inter-regional industrial structure and trade
relationships play a critical role in determining the e ects of various policies and programs. To underscore
this proposition, we turn our focus to the commonly pro ered argument that there is an inevitable trade-
o between national e ciency and interregional equity, and that to enhance the latter requires sacri cing
the former. Using this simple numerical example, we demonstrate that it is indeed possible, as suggested
in Section 2.2, to enact policies that can target economic enhancement in lagging regions in ways that
not only do not sacri ce national e ciency, but actually enhance it.
Our policy scenario is as follows. Assume that we have a national budget to support expenditures in the
amount of $90k. Such an expenditure of public capital might be channeled to physical infrastructure, to
educational infrastructure, or even direct assistance. Now, because productivity as measured by value
added per capita (PCVA) is highest in Region I, conventional wisdom might suggest that focusing the
public capital expenditure in the most e cient region would have the greatest impact on aggregate e
ciency, as measured by increase in value added. To demonstrate that this is not a foregone conclusion,
we use three scenarios, one that represents each extreme position, with all $90k invested in Region I or
Region III, and an intermediate scenario in which each region receives equal shares of the public capital
investment, as shown in the table 4.

13
 Table 4: Scenario Development
Regional Expenditure
Distribution:
Scenarios A, B, and C
A B C
I 90 30 0
II 0 30 0
III 0 30 90
The impacts assessment outcomes are shown in table 5 below. As expected, value added impacts are
greatest in the investment regions for the two extremes, and when the investment is spread evenly over
all three regions, the impacts are directly correlated with pre-shock development levels. Table 5: Impact
Results
VA Impacts by Scenario PCVA Impacts by Scenario

Region Popn VA PCVA A B C A B C

I 10 450 45.0 60.6 25.5 7.8 6.1 2.8 1.0
II 9 308 34.2 3.9 23.0 7.6 0.4 2.6 1.0
III 8 258 32.3 4.2 20.9 54.7 0.4 2.3 6.8
Nation 27 1016 37.6 68.79 69.43 70.07 2.29 2.57 2.92
However, contrary to conventional wisdom, the distribution of public capital investment that generates
the largest value-added impact is Scenario C, in which the entire investment is targeted to Region III, the
one with the lowest PCVA. This also is the scenario that results in the strongest convergence in PCVA
across regions, and the greatest national PCVA value (see table 6).
Table 6: Modi ed Results
New PCVA by Scenario Percent Change in PCVA

Region Popn VA PCVA A B C A B C

I 10 450 45.0 51.1 47.6 45.8 13.5% 5.7% 1.7%
II 9 308 34.2 34.7 36.8 35.1 1.3% 7.5% 2.5%
III 8 258 32.3 32.8 34.9 39.1 1.6% 8.1% 21.2%
Nation 27 1016 37.6 39.50 39.7 39.98 5.0% 5.6% 6.2%
3
6.9 10.1 6.8 5.4

The lessons that we draw from this exercise can be summarized as follows:
1. The structure of interindustry interregional trade within a nation plays a vital role in determining
impacts of policies and programs on each region and on the nation.
2. Investments in one region will have impacts on all regions.
3. Whether a nation must sacri ce e ciency in exchange for increased interregional equity will depend
upon the intra-and inter-regional interindustry structure of that national system.
4. Altering the structure of intraregional trade can have dramatic impacts on own-region multipliers
and system-wide e ects.

14
To further illustrate this nal point, consider the following scenario in which, as a result of import
substitution, Region III increases its intraregional dependency, as re ected by an increase in its diagonal
value from 0.28 to 0.31. Making this single change while holding all else constant in Scenario C results in
a 4.4% increase in both the regional and national value-added impacts values. Likewise, strengthening
interregional trade can have substantial impact. Doubling Region I’s relatively small purchases from the
other two regions increases the value-added impact by 1.75%.
Together, these examples indicate that a) a comprehensive understanding of the geographical
distribution of national policy impacts requires an understanding of interregional inter-industrial
structure. The positive impacts of policies and programs can be targeted to speci c regions, and relative
distribution of bene ts across regions can be a ected by careful structuring of national programs.
Therefore, there is great value in developing a systematic approach to the altering the characteristics of
regional structure that can have the greatest e ect on regional and national development. The analysis of
regional industrial structure is the topic of section 4.
The purpose of the example in this section has been to demonstrate that there is a variety of outcomes
that are possible given alternative public capital investment scenarios and public policies. It is also
undoubtedly clear that the speci c outcomes that will be observed depend heavily on the detailed
structure of regional and interregional economies. Every summary value in the example scenario
embodies a number of characteristics and details that would require much greater speci cation and
accuracy for any real-world application. Di erent industries have di erent input requirements and output
distributions, they have di erent ties of di erent strengths to other regions, and their value-added
payments are distributed di erently, not only among the major categories of compensation, payments to
governments, and gross operating surplus, but also to di erent household categories. Labor force
requirements create di erential demands on households of di erent types and income levels, and di erent
demographics. And these are just some of the many detailed characteristics that require explication
when analytical models are constructed for assessing alternative policy outcomes.
Additionally, moving beyond those regional and interregional characteristics that can be captured in
interregional accounting frameworks, other regional traits and behavioral di erences can and do
moderate the ways in which policy outcomes play out from region to region. Di erences in natural
resource endowments and amenities often de ne and constrain the development path. Di erences in
regional consumption savings rates, for example, can have a dramatic in uence on the ow on e ects that
play a critical role in determining the total impacts of policies. Likewise, whether savings translates to
investment within or outside a regions’ borders will strongly shape the developmental impacts of
increased income, savings, and investment.
Other factors that will shape policy outcomes include the extent to which development may be aided by
external economies of scale and agglomeration, by cultural norms that will in uence educational and
hence occupational skill levels, labor force participation rates, family structures, the prevalence of crime
and other social maladies that often arise from increasing density and congestion.
In the next section, a more comprehensive review of some regional analytical methods will be provided,
after which attention will be directed to methods of exploratory spatial data analysis.

15
4 Regional Economic Analysis Methods: Introduction and Selective Overview
At the most fundamental level, a complete description of regional economic structure begins with an
inventory of employment, compensation (wages and salaries), and output by industry. This information
contributes to economic understanding in at least four ways. First, it is useful to know how specialized or
diversi ed the regional economy is. A long-standing concern at the regional level is overspecialization
that makes a region susceptible to industry-speci c national and global economic downturns. Consumer
tastes and preferences and industrial production needs shift over time, so regions that are too highly
specialized can su er traumatic consequences if industries on which they depend experience rapid
declines in demand. Second, by identifying relative concentrations by industry, we can begin to assess
regional comparative advantages. Industries that fall into this category become candidates for further
analysis, assessing their futures and potential for further development, not only within these speci c
industries, but also in related industries that are strongly connected by direct and indirect forward (sales)
and backward (purchases) linkages. Third, a comprehensive inventory of activities becomes the
foundation for behavioral models that can be used for assessing economic impacts of policy changes,
positive or negative economic disruptions, and consequences of public capital investments. These
models can be used to anticipate and understand regional economic system responses in terms of
changes in employment and wages, occupational demand, and changes to household income
distributions. Fourth, models of individual regions can be linked with other regional models to develop a
better understanding of how regional fortunes are intertwined.

4.1 Regional Economic Systems Inventorying

In this section, we provide a brief overview of a set of fundamental methods that most often form the
initial assessment of regional economic structure. The rst set largely composed of quotient-based
metrics focuses solely on description, while the second set lays the foundation for prospective
interventions by incorporating either a temporal or behavioral element.

The Location Quotient

The location quotient, or LQ, is a very simple metric that is designed to compare a distribution of shares,
or concentrations, with a reference distribution. In regional analysis, it has most often been used to
assess concentrations of industries by focusing either on regional employment or regional income by
industry relative to the national distribution. In this case, the industry share of national total is used as a
meaningful comparator for subnational regions, so industry LQs can be directly compared across regions.
An example would be to compute LQs for variable x (e.g., employment or income in industry i) as

region’s share of national xi regional xi/x

or
region’s share of national x national xi/x

Using employment as the interest variable, the rst expression carries the interpretation of the region’s
share of national employment in a speci c industry relative to the region’s share of total employment,
and the second expression is expressed as the industries share of regional employment relative to the
region’s share of national employment. Although the objective for a given analysis might lead to a

16
preference for one or the other expression, their mathematical equivalence is assured, as is their
interpretation. LQ values greater than 1.0 indicate relative regional industry concentrations, and values
less than 1.0 indicate concentrations smaller than the national average.
More formally, in terms of employment, the location quotient for industry i is de ned as

(1)

where ei denotes regional employment in industry

i, Ei is national employment level in industry i,
e is total regional employment,
E is total national employment.
While there can be many explanations for high and low concentrations of industries in a given region,
high LQs suggest regional comparative advantage for the industry, and therefore they point to industries
that might be important exporters and sources of regional product, and low LQs suggest that these
industries might be underrepresented in a region and hence their products might need to be imported,
hence accounting for some negative entries in regional balance of payments. These speculations depend
on a range of assumptions about industry mix, consumer tastes and preferences, and the like, so LQs are
clearly only broad-brush indicators, but can be useful in pointing the analyst to industries that warrant
further attention. Other methods discussed below will relax some of the assumptions that underpin LQ
interpretations.

Related Measures of Concentration

A related measure that focuses on a single industry’s concentration across all regions is the Coefcient of
Localization (CL). The CL relies not on the ratios of ratios, but instead is computed as one-half the sum of
the absolute di erences of between numerators and denominators in the LQ calculations for the industry
in all regions. The closer this value is to unity, the more concentrated the industry is in the nation. An
example for a ctitious industry and nation can be seen in the table below, where the LQ indicates that
the industry is relatively concentrated in regions 4 and 5, and the value of the CL is 0.21. Industries with
low CL might be less tied to speci c regional attributes, and therefore might be good candidates for
regional diversi cation in regions where the LQ is less than one. Diversi cation strategies will be discussed
further, below.
All Regions

Region 1 Region 2 Region 3 Region 4 Region 5 Region 6 Total

A)Region’s share of
national industry 0.050 0.070 0.150 0.250 0.290 0.190 1
employment
B) Regions share of
national total 0.100 0.150 0.200 0.150 0.180 0.220 1
employment
|A - B| * 0.5 0.025 0.040 0.025 0.050 0.055 0.015 0.21
LQ 0.500 0.467 0.750 1.667 1.611 0.864

17
More formally, the CL equation can be written as

(2)

LQs and CLs also are related to Lorenz Curves, which graphically display the unevenness of distributions,
and can be used to compare industry distributions by plotting more than one Lorenz curve on a single
graph. To generate the Lorenz curve for an industry, regions are ranked by location quotient, and
regional shares of national industry and national total employment are plotted on a cumulative basis, as
shown below.3 The slopes of straight-line segments are location quotients of regions. The ratio of the
area between the curve and the diagonal line to the total area of the right triangle is another coe cient
of localization. The closer this ratio is to 1.0, the more uneven is the distribution.

Industrial Diversity
Until relatively recently, industrial diversi cation was a dominant regional economic development
strategy. The diversi cation strategy rests on the recognition that while a heavy concentration of
employment in a single industry or set of closely related industries might carry some advantages, which
will be discussed in the context of industrial clustering as a development strategy, it also carries with it a
concentration of risk should the industry experience a downturn. By diversifying a regional economy, the
region would be diversifying the risk of employment downturn in the same way as an investor guards
against heavy losses by diversifying her stock portfolio. Diversi cation, then, would be expected to
contribute to employment stability.
Despite the more recent emphasis on potential economies of industrial clusters and implications for
more rapid growth, industrial diversity remains an important indicator of economic structure. Whereas
3 When the order of industries is reversed, the Lernez Curve will lie below the diagnonal.

18
the LQ and related concentration measures focus on individual industries, there is another set of
measures that are used to quantify the degree of specialization or diversity of industry structure for all
industries in a region. Among the most common regional industrial diversity indices are the National
Average Index, the Her ndahl Index, and the Entropy Index. Each of these metrics provides a summary
measure of the di erence between a given region and its national economy in terms of industrial
distributions. The implication is that the national economy itself is in some sense optimally diversi ed,
hence these indices become normative. Of course, it would be possible to substitute any alternative
industrial distribution for the national economy if there were rational economic reasons for doing so.
These measures are also usually based on income or employment statistics, largely because these data
are typically more readily available and reliable than estimates of output. However, any of these
measures could be used.

Common Diversity Indices

Each of the industrial diversity indices rests on a normative assumption concerning the ideal regional
industrial economic structure. Because the normative ideal can be expressed in a variety of ways, the
measures di er accordingly. The national average index is among the earliest and simplest measures of
regional industrial diversity. It emerged in the context of highly developed economies, where the
national economy could be assumed to be diverse, and its structure taken as the norm. A perfectly
diverse regional industrial structure would therefore mirror the national industrial structure, and
departures from the national industrial concentrations re ect regional specializations, or departures from
perfect diversity. When deviations are expressed in absolute value, the virtually identical to the coe cient
of localization. When deviations are expressed as squared di erences, the equation for the National
Average index for region R is

(3)

A value of zero re ects perfect industrial diversi cation and larger departures from zero re ect greater
regional industrial specialization.
A second regional industrial diversity index that has become increasingly popular in the related literature
is the Her ndahl-Hirschman Index (HHI). This measure, which also is often used in studies of industrial
organization and market power, is a function only of the region’s own industrial structure, rather than a
relationship to a reference region. The normative ideal in this conceptual framework is equal shares of
employment (or income or other industry size metric) in all industry sectors. A perfectly diverse region
would thus have 1/N of its employment in each of its N industries. Departures from the normative 1/N
share in uence the HHI strongly, as the industrial shares are squared are summed, as shown in the formal
equation below.

(4)

The HHI value can vary from a minimum of 1/N to a maximum of 1.0. The maximum value would indicate
that the region has only one industry.

19
A nal index of diversity is derived from information theory and is known as the entropy index. The norm
with the entropy measure is similar to that of the HHI, but this measure is sensitive to the number of di
erent industries in a region, in that its maximum value increases as that number increases. The form of
the expression most often use is referred to as Shannon’s H, the formula for which is

(5)

Shannon’s H values range from 0 to lnN . Because this measure is sensitive to numbers of industries, its
value re ects not only the distribution of industry employment (or other relevant) shares, but also the
numbers of industries that compose the di erent regional economic structures. This can be useful for
monitoring intertemporal change, but for comparisons across regions, the value is often normalized by
its maximum, to once again yield values that range from a minimum of 1/N to a maximum of 1.0. The
normalized equation is

(6)

Although industrial diversity continues to be a mainstay of regional economic structural assessments,

another side of the economic development coin focuses on the bene ts of key industries, and clusters of
similar and synergistic industries.

Key Industries
A prominent line of reasoning that can be traced growth pole/growth center theory revolves around the
concept of key industries. Following work by Hirschman (1958) and Rasmussen (1956), whose measures
were founded on input-output (IO) relationships among industries, key industries became the focus of a
large literature. IO facilitates the identi cation of industries that exert greater than average in uence on
their economies by virtue of their forward (sales) and backward (purchases) linkages with other
industries. Intermediate sales and purchases bind industries one to another and create
interdependencies. Those that occupy the most prominent positions in multiple supply chains are identi
ed as key industries. The absence of such industries in a region, where these industries are deemed to be
capable of operating economically, have become the basis for regional public agencies’ e orts to attract
these key industries.4

4.2 From Individual Key Sectors to Industrial Clusters

Many regions’ policymakers have abandoned development policies focused on individual sectors (such
as key sector strategy) in favor of a cluster-based approach. Rather than focusing on individual key
sectors, this approach centers on the identi cation of clusters of industrial sectors with substantial
mutual synergies based on mutual sales and purchases interdependencies (Feser and Bergman, 2000).
These strategies, which rest on the theory of agglomeration economies, emphasize ways of promoting
and supporting these clusters to facilitate and deepen their development. Bekele and Jackson (2006)
have identi ed and discussed the most common approaches to industrial cluster identi cation.

4 For a recent review and discussion of key sector measures, see (Temurshoev and Oosterhaven, 2014).

20
Melding Clustering and Diversi cation Strategies
Despite the apparent contradictions between clusters and diversity, however, there has been increasing
discussion of the coexistence or at least creative perspectives on clustering and diversi cation.
Jackson (2015) has provided an input-output based method for devising strategies that recognize
comparative advantages in existing regional structures, quantify the strength of clusters already present,
identify gaps and bottlenecks in cluster supply chains, explore the regional consequences of potential
cluster diversi cation strategies that might serve to further diversify regional industrial structures.
Outcomes from such strategies have been called diversi ed clusters. Jackson’s cluster assessment diversi
cation strategy (CADS) follows these general steps:
1. Use a single-sector method to select an industry or set of industries from a study region for
consideration as the anchor industry or industries of one or more clusters,
2. Identify the current employment or output levels for these anchors,
3. Determine the industrial distribution of supporting, supply chain-linked industries that would be
needed to fully support these anchors at existing production levels,
4. Assess the su ciency of the anchor industry or industries for supporting the selected clusters by
comparing existing production to supply-chain requirements, and
5. Evaluate the implications of diversifying into new specializations.
CADS provides a set of detailed instructions for implementing these steps, which can identify existing
strengths and gaps or bottlenecks in the regional economy in terms of supply de cits; it can be used as a
standard against which the existing distribution of industrial activity can be measured; and it can be used
to identify the distributional implications of diversifying into new industry clusters, all of which will can
lead to more rational economic development decisions (Jackson, 2015, p. 123).

4.3 Identifying Supra-Regional Trends

Nearly all of the metrics discussed thus far are founded on static concepts. As such, there is an
unintended and potentially misleading implication that all industries are equally viable candidates for
development, from both regional and supra-regional perspectives. This conclusion could lead a regional
policymaker to work toward the recruitment or promotion of a targeted industry that simply might not
be competitive in a given region. One broad brush approach that has allows the decomposition of
changes in industrial structure into components that re ect national (supraregional) trends, industry mix,
and region-speci c characteristics is called Shift-Share Analysis. This method has been in use for decades
and owes its longevity to simplicity of implementation, interpretation, and extension. 5 Shift-Share
Analysis has been used in a variety of policy contexts, including the identi cation of regional comparative
advantages and other region-speci c, relative strengths and weaknesses.

Shift-Share Analysis6

5 More sophisticated methods called structural decomposition analyses have been developed using input-output accounts
as empirical foundations. See Lahr and Dietzenbacher (2017) for details.
6 This section draws heavily on Jackson and Haynes (2009), where various issues involved in selecting an appropriate time
period for analysis, data development, and methodological extensions are addressed.

21
Once again, we use regional employment by industry as our indicator of industrial structure, although
personal income or output data by industry would be equal informative when these data if and when
these data are available. This method, however, requires industrial structure data for two di erent time
periods. While there are numerous extensions, including dynamic shift-share (Bar and Knight, III, 1988)
and the incorporation of spatial structure (Nazara and Hewings, 2004), the presentation below re ects
the method in its most fundamental form.
With data for two di erent time periods, we use time t−l as a reference to the beginning period and time
t to the ending period. Our industrial structure changes will thus refer to the period spanning time t−l to
time t. Our variables are de ned as follows: ert - total employment in region r at time t eri,t - employment
in region r industry i at time t ent - total national employment at time t eni,t - national employment in
industry i at time t We rst calculate the following three growth rates:

= growth rate for regional industry i (7)

= growth rate for national industry i (8)

= average national total growth rate (9)

By de nition:

(10)

If regional industry i grew at the average national rate, its change over the time period would be

eri,t = neri,t−1 (11)

and the change in industry i in region r would be attributed solely to average national trends. However, if
industry i is particularly fast or slow growing, there will be a di erence between n and ni. We can account
for the deviation of a given industry from the national average using the following expression:

eri,t = eri,t−1[n + (ni − n)] (12)

If the industry is fast-growing nationally, eri,t will be greater than the value obtained from equation (11),
and it will be less than that value if it is growing more slowly than the nation.
If industry i grows at di erent rates in di erent regions, then there will be a non-zero di erence between ri
and ni . We can capture this e ect with the following expression,

eri,t = eri,t−1[n + (ni − n) + (ri − ni)](13)

In which the bracketed term reduces to , which duplicates the identity in equation (7).

22
The integrity of the accounting system that de nes changes in regional employment by industry can be
veri ed by subtracting eri,t−1 from both sides, yielding
eri,t − eri,t−1 = eri,t−1[n + (ni − n) + (ri − ni)] − ei,tr−1 (14)

or
∆eri = eri,t−1[n + (ni − n) + (ri − ni) − 1] (15)

and
∆eri = eri,t−1[(n − 1) + (ni − n) = (ri − ni)], (16)

where ∆eri is the study period change in regional industry i employment.

By separating the components of the right-hand-side of equation (16) we can identify the following
values and interpretations:

The national share component Ni = eri,t−1(n−1), which is the growth in the regional industry that can be
attributed to the national average rate of growth;

The industry mix component Mi = eri,t−1(ni−n), which identi es the growth in the regional industry due to
the di erence between the speci c industry’s national growth rate and the national average rate of
growth; and

The regional shift Ri = eri,t−1(ri − ni) regional industry i growth that cannot be attributed to industry or
national e ects. This regional shift component is often used as a measure regional comparative
advantage for industry i .
Given these de nitions,
∆eri = Ni + Mi + Ri (17)

Finally, we can sum both sides over all industries and show that

X r X
e
∆ i= (Ni + Mi + Ri) (18)
∆er = N + M + R (19)

Shift-Share Analysis provides results for each industry that yield speci c interpretations. Combinations of
negative and positive values for the Mix and Regional Shift components characterize the industry’s
performance relative to the national average industry, and its performance in the region relative to its
performance in other regions. These are summarized in the table, below.
M RS Interpretations
- - The industry was a poor performer nationally, and even poorer in the region
than in the nation.
- + The industry was a poor performer nationally, but better in the region than in
the nation.
+ - The industry was a good performer nationally, but poorer in the region than in
the nation.

23
 + + The industry was a good performer nationally, and even better in the region
than in the nation.
This summary information can be used to supplement and reinforce (+ +) or dampen (- -) regional
enthusiasm for industries that other methods might identify as key industries or industry clustering
targets.
4.4 Industrial Restructuring Challenges
Virtually all of the industry targeting and industrial restructuring methods must be considered in the
context of local understanding and more comprehensive knowledge of the study region. These strategies
can be complicated by a number of considerations, some of which are listed below.
1. Changes in rm ownership patterns can result in organizations of production that align more
strongly with aspatial competitive advantages rather than with locational advantages. Some
analysts have questioned whether geographical proximity is the major source of agglomeration
economies.
2. Missing supply chain linkages might re ect the lack of a su cient volume of demand to allow the
activity to reach its minimum e cient scale, and hence, make that sector viable in the region.
3. Decreasing real transportation costs allow rms to search over wider geographies for inputs and
markets.
4. Labor force skills and quality, physical infrastructure, and other industry-speci c requirements must
be su cient to support targeted industries.
The ability of regional policy makers to e ect signi cant change is in many cases constrained by national
policy and global forces. Accordingly, attention is often directed to changes at the margin enhancing a
region’s connectivity, investment in human capital and the provision of incentives or indirect support to
enable local rms to upgrade technologically. Much less attention has been given to addressing the
market failures that are in the scope of most regional and local governments, such as ensuring the uidity
of land markets or promoting a favorable business environment. The recent attention directed to the
idea of smart specialization provides an example of an innovation in policy thinking that has been clearly
motivated by the di culties of measuring the contributions of past policy initiatives.

5 Regional Accounts and IO Software

Regional Accounts
Regional accounts are to regions what national accounts are to nations. These accounts provide the
foundations for a wide array of analytical tools and techniques. The accounts detail the interactions
among economic agents within the system and with other regions and the rest of the world. These
economic agents are the establishments and rms that compose the production sector of the economy,
and the consumers of the goods and services that they produce, which includes households,
governments, investment, and trade. The establishments are grouped in industries, and their products
are grouped into corresponding commodities.
Because region economic systems are typically more open than national counterparts, an accurate identi
cation of which interactions take place within the region and which cross the regions borders in the form
of imports and exports is critical for the generation of meaningful analyses using the suite of tools and

24
techniques that are available. Analysts’ initial challenge is often the acquisition or construction of
regional accounts. This challenge is ampli ed by two factors. First, regional accounts constructed from
primary data are rare. Most regional analysts must rely on one of several methods that can be applied to
generate regional accounts from corresponding national accounts using a smaller set of region-speci c
data to estimate counterparts to national parameters. Second, whereas most of the tools that have been
developed for national or regional input-output accounts operate on industry-by-industry (or
interindustry) accounting frameworks, it is a reality of the data collection and reporting conventions that
most nations publish accounts not in interindustry format, but in what is known as a supply-use
framework. Interindustry accounts can be derived from supply-use data, but the necessary data
manipulations are not always straightforward and intuitive.
As an aid to regional analysts, computer software applications have been developed primarily for the
purpose of facilitating the construction of regional accounts. For example, IO-Snap, described further
below, supports accounts generation and standard nal demand impacts assessment, but more extensive
analytical capabilities beyond multiplier analysis, descriptive economic statistics, and impacts
assessment are still being added. While additional analytical features are under development, however,
IO-Snap can be used to generate regional interindustry accounts that can then be exported for use with
software applications like REAL I-O that provide a much more comprehensive suite of analytical tools. IO-
Snap and REAL I-O are described brie y, below.

IO-Snap
IO-Snap, which stands for Input-Output State and National Analysis Program, is a Windowsbased
computer software program that was developed to facilitate the use of input-output data from U.S.
national make and use tables. In the process of assembling the various utilities that support national and
state-level input-output analysis, the developers also created an environment that facilitates inter-state
comparisons of input-output related variables such as employment and worker compensation. Annual
U.S. IO tables from the Bureau of Economic Analysis (BEA) for 1998 through the most recently published
year are included with the software. In addition to default data covering the U.S. 50 states and the
District of Columbia, user-supplied data in make-use format can be imported for other geographic areas.
Make and use tables contain a wealth of data in their original formats. Production functions and output
distributions, and nal demand activity supplemented by employment and compensation by industry data
can be easily accessed and extracted for use in other applications. Using the default state-speci c data or
user-supplied data for other U.S. regions, users can generate regional input-output accounts. With the
national, imported, or user-generated tables, users can create Industry by Industry, Industry by
Commodity, and Commodity by Commodity direct and total requirements tables for a variety of
supported applications. Final demands by commodity or by industry can be speci ed, and impacts
assessments results can be produced. Standard results can be generated, including income and
employment impacts, disaggregated and total multipliers. Cut and paste options are enabled throughout
so that users can easily transfer data to spreadsheets or other analytical software for further analysis,
graphing, etc.
Within IO-Snap, data can be edited and displayed in a variety of formats, compared and contrasted
across geographical de nitions at di erent times, and fundamental input-output based analyses can be
implemented. Prior to generating the requirements tables, the user can modify use table columns
(including nal demand distributions) and make table rows. Available codes of table operations and
analysis are outlined in the Annex. IO-Snap software is available from [Link]

25
REAL I-O Software
REAL I-O, an input-output operation software is a generic toolbox of Input-Output (IO) analysis based on
open-source architecture running on Windows XP/7. Following the previous versions, termed PyIO
(Nazara and Hewings, 2004), Python is retained as the interface building software.
However, the main modules of matrix calculations have been currently migrated to the R language
environment (SPlus equivalent freeware). This change allows the users to introduce their own database
and additional functions in a much more convenient way than in previous versions. The migration will
continue over the next several months and should be complete by the end of August 2018.
At the moment, several analytical functions of intra and inter-regional input-output analysis are
preloaded in the REAL I-O package. The available codes of table operations and analysis are listed in the
Annex. The latest version of the REAL-IO software is available at [Link] [Link]/realio/. The
example data sources (e.g., OECD STAN Input-Output Database for 44 countries and Inter-country inter-
industry) are also included.

6 Spatial Analysis of Regional Economies

Regional economics o ers a rich empirical setting for the application of geospatial methods, as our review
of regional economic development research has highlighted a number of theoretical constructs that call
out for empirical measurement. In this section, we outline a number of such methods, linking each to
particular theoretical concepts and discuss their implementation and use in the analysis and monitoring
of regional economic systems.
We begin with measures of inequality, polarization and concentration for a national system composed of
interacting regional economies. Closely related to, but distinct from, polarization measures are
measurements of spatial clustering, which include global measures of spatial autocorrelation. We
provide an overview of so-called global measures of spatial autocorrelation as well as their combined
use with measures of polarization. We then shift the spatial focus to discuss local measures of spatial
association. These provide the ability to detect hot(cold)-spots of economic activity that might be
responsible for the overall patterns detected by the global measures or spatial-outliers that depart from
the global pattern.
The measures of polarization, clustering, and local spatial association are all focused on static patterns
measured at one point in time. Recent work in spatial statistics and GIScience has focused on developing
new methods that extend these to consider the dynamics of these patterns. Methods of exploratory
space-time data analysis can be used to study questions of spatial income mobility that has to date been
largely absent from the debates surrounding the regional equity-e ciency tradeo . We close this section
with an overview of recent developments of machine learning in GIScience which may o er new
approaches to the study of regional economies.

6.1 Inequality, Lorenz Curves and Convergence

Inequality for a collection of R regions in a national system has been measured in a number of ways. The
concept of σ -convergence Young et al. (2008) is perhaps one of the most widely employed, relying on
the sample standard deviation:

26
 (20)

where yr,t is income in region r at time period t and For comparative analyses, the coe
cient of variation is sometimes employed:

(21)

These measures of σ-convergence are scalar indicators that capture one aspect of the entire distribution
of incomes across the R regions in the national system. Measures that are designed to consider more
fully the distribution can be obtained by repurposing the Lorenz Curves and Gini coe cients, we
previously encountered in the study of regional industrial structure, by replacing measures for di erent
industries within a region with say per-capita incomes across the R regions (Frick and Goebel, 2008).
Polarization within the interregional system has been measured using regional adaptations of the
Esteban and Ray (1994) index
R R

p(α)t = XXs1+r,t asq,t|yr,t − yq,t| (22)

r q

where α is a parameter that expresses the sensitivity to polarization, and sr,t is the population share of
region r (Ezcurra, 2009).

6.2 Spatial σ-convergence and Spatial Gini

The previous measures of inequality and dispersion are silent on the geographical arrangement of the
regional economies. Put another way, in addition to the observed map consisting of R regions, there are
a total of R! map patterns that could be generated such that each map obtains the same values for the
measure of inequality at hand. Because of the locational invariant property of traditional measures of
inequality, several authors have suggested spatially explicit measures for σ convergence (Rey and Dev,
2006; Egger and Pfa ermayr, 2006). These allow for the partitioning of overall dispersion/inequality into
a pure inequality component and spurious dispersion that arises from the complications due to spatial
dependence. In the same spirit as spatial σ-convergence, Rey and Smith (2013) have suggested a spatial
decomposition of the Gini coe cient which distinguishes between pairwise inequality for neighboring
economies versus inequality between pairs of regions that are geographically distant from one another.

6.3 Theil Interregional Inequality Decomposition

R

X
Tt = strlog(Rstr) (23)
r=1

and:
R

str = yrt/Xyrt (24)

27
 i=r

where R is the number of regions and is per capita income in region r in period t.
Decomposition of total inequality is obtained as:

) (25)
where ng is the number of observations in group is the share
of total income accounted for by group g, and is region i’s share of group g’s
income.
Regional applications of this decomposition are based on an exhaustive and mutually exclusive
assignment of regional economies to groups. As a result, inequality is split into that due to average di
erences between regional groups (so called interregional inequality, the rst term) and inequality
between regions belonging to the same group (intraregional inequality, the second term). Shorrocks and
Wan (2005) summarize the literature nding that the interregional component is often smaller than the
intraregional counterpart.

6.4 Exploratory Spatial Data Analysis

Recent developments in the eld of exploratory spatial data analysis (ESDA) a ord numerous opportunities
that can be used to analyze interregional linkages. The exploratory nature of these measures implies that
their role is to characterize spatial patterns in terms of departures from spatial randomness due to
spatial clustering and or dispersion. By themselves, these methods are not intended to explain these
patterns but are powerful diagnostics to detect spatial structure and suggest that spatial processes are at
work. This exploratory stance is sometimes dismissed as measurement without theory , however we feel
that this critique is misplaced. Indeed as Rey and Le Gallo (2009) argue, ESDA methods can o er
important complementarities to formal econometric modeling of regional economic systems. We see
much potential for a more comprehensive analytical framework for regional economic analysis that
integrates ESDA methods with economic process modeling including spatial econometric, computable
general equilibrium models and interindustry modeling frameworks of input-output and social
accounting

6.5 Global Autocorrelation Measures

Spatial autocorrelation is an articulation of Tobler’s rst law of geography: Everything is related to
everything else, but near things are more related than distant things (Tobler, 1970). From an economic
perspective, the presence of spatial autocorrelation can re ect the operation of many processes that link
regions together. Labor migration, capital ows, trade networks, value chains, intergovernmental transfers
and policy copy-catting are but a few of such processes that can give rise to spatial association.
It is important to note that autocorrelation and concentration, while similar, actually capture di erent
characteristics of economic space. Concentration or polarization derives from the a-spatial attribute
distribution - and speaks to the degree of modality. A high degree of modality would indicate that the
attribute in question (say regional GDP) is concentrated in sets of regions but does not necessarily
provide any further information about the spatially explicit nature of those sets of regions.

28
6.6 Local Autocorrelation Measures (Spatial Clusters/Hot-Cold-Spots)
The global measures of spatial autocorrelation provide what are referred to as whole map statistics. That
is, they provide scalar measures of the extent to which the map pattern as a whole departs from the null
hypothesis of spatial randomness. Important complements to the global measures are local indicators of
spatial association (Anselin, 1995). These serve two main roles. As focal measures, the LISAs can be used
to identify so called hot (cold) spots that may be driving the overall pattern of spatial association. For
example, a map with strong positive global autocorrelation might re ect patterns of value-similarity in
space (neighboring units having similar attribute values), while the LISA values for each region can be
inspected to identify the relative importance of local units for contributing to the global pattern.
The second role for LISAs is to detect spatial outliers. These are local units that display association
distinct from the overall pattern. For example, in the case of positive global spatial association, a local
spatial outlier would re ect negative association with its own attribute value being inversely related to
those of its neighbors. These locations could become the focus for subsequent investigation or targeting.

6.7 Markov and Spatial Markov/Distribution Dynamics

The measures of inequality and polarization reviewed above focus on a single cross-section of regional
economies. The evolution of these measures can provide policy makers insights as to the e ectiveness of
policies designed to reduce regional disparities. Yet these measures are fairly aggregate in scope,
informing on the external characteristics of regional income distributions such as whether polarization
(modality) or dispersion (sigma convergence) are declining or worsening.
The rise of the distributional dynamics school of analysis can be seen in part as a response to this
external focus, as well dissatisfaction with the overly restrictive nature of formal models of economic
convergence (Fingleton, 1999). Distributional dynamics has introduced the concepts of discrete Markov
chains (Quah et al., 1993) and stochastic kernels (Quah, 1996) to study the internal dynamics of regional
income distributions. More speci cally, these approaches can quantify the amount of internal mixing and
churn in regional income distributions over time, as well as support measures of regional income
mobility, a topic we return to below in more detail. Closely related to stochastic kernels is the application
of stochastic dominance to study the evolution of regional income distributions over time (Carrington,
2006).
Stochastic kernels and discrete Markov chains have seen widespread application in the regional
economics literature, following their use in international comparative studies. However, their application
at the regional scale had largely ignored the empirical regularities of regional income series, namely
strong levels of spatial dependence and/or spatial heterogeneity. Several extensions of the traditional
Markov framework have been suggested to incorporate spatial dependence. The spatial Markov
framework (Rey, 2001; Rey et al., 2016) allows for the transitional dynamics (i.e., the probability of a
region moving up or down the regional income distribution) to be related to the levels of income found
in its neighboring regions. Application of spatial Markov methods has uncovered widespread evidence of
spatial poverty traps and growth magnets (Le Gallo and Ertur, 2003; Mossi et al., 2003; Bosker, 2007;
Bosker and Krugell, 2008; Villaverde and Maza, 2012).
The Markov paradigm has also provided a mechanism to move the cross-sectional measures of spatial
autocorrelation into a dynamic frame. By embedding the quadrants of a Moran scatterplot (HH, LH, LL,
HL) as states in a Markov chain, a LISA Markov can be developed which a ords a rich taxonomy of space-
time transitions (Rey and Janikas, 2006). The following table reports results from application of the LISA

29
Markov to 32 Mexican states over the period 1940-2000. The transition matrix is diagonally dominant,
with the HH and LL states displaying the strongest staying probabilities. In the long run, this is estimated
to result in a strengthening of spatial clustering as the ergodic distribution accumulates more mass in
these two states.

Ergodic Distribution 0.326 0.139 0.373 0.161

6.8 Mobility
Although regional inequality and growth have commanded the vast majority of attention in the regional
science literature, the concept of economic mobility has recently begun to draw focus. Regional
scientists are borrowing from the literature on intergenerational income mobility (Maasoumi, 1998)
where the focus is on the upward or downward movement of individuals in an income distribution
relative to the position of their parents. A spatial turn to the notion of income mobility replaces the life-
time earnings of parents and their o spring with regions at di erent moments of time and considers how
regions may, or may not, change position in the regional income distribution. As mentioned above, the
Markov framework can be used to provide aggregate measures of spatial income mobility. More
recently, there have been a number of new measures of spatial income mobility employing concepts of
rank concordance and circular statistics (Hammond and Thompson, 2002; Rey et al., 2011; Rey, 2014,
2016).

6.9 Integrating equity, e ciency and mobility

There is a need to revisit the debate surrounding regional inequality and national growth (e ciency) to
introduce mobility (Moser and Schnetzer, 2017; Rey, 2018). It is generally acknowledged that studying
inequality or national growth separately provides only a limited understanding of either phenomena.
The same argument holds for the concept of regional income mobility. For example, estimating the rate
of spatial income mobility, while important may fail to address broader questions about the role of that
mobility in interregional welfare. In the intergenerational literature, much attention has been placed on
the question as to whether mobility is equalizing in long run incomes (Fields, 2010). We currently do not
know whether regional income mobility works to reduce regional disparities, nor do we know the
relationship between national economic growth and spatial income mobility. This tri-variate relationship
will require new models and frameworks but o ers the potential for a more holistic understanding of
regional development., ,

6.10 Spatial Counterfactuals

Counterfactuals provide a way to disentangle the role of changes in attributes and the changes in model
parameters as drivers of distributional change. (Fingleton and Palombi, 2013; Carrillo and Rothbaum,
2016). Originally used in the analysis of gender discrimination in the labor market, the approach relies on
the construction of two di erent conditional wage distributions. The rst is estimated for female
participants in the labor market where the wage distribution is conditioned on observed years of
schooling and experience. In a similar fashion, a conditional distribution is also estimated for male
participants. A counterfactual distribution is then estimated for females re ecting the conditional wage

30
distribution that would obtain given their distribution of experience and schooling and the returns to
these characteristics for male labor market participants.
Carrillo and Rothbaum (2016) have turned a spatial lens on this approach in the context of modeling
urban residential location choices. The approach rst estimates conditional distributions for the
residential location of Hispanics in Washington D.C. subject to observed individual characteristics
including age, education, and gender. With this conditional distribution in hand, the change in the actual
distribution of residential locations of Hispanics over an interval is decomposed into components due to
changes in the returns to these characteristics in in uencing locational choice, and changes in the
characteristics themselves over the two periods. This is accomplished through the creation of the
counterfactual distribution for residential locations in the second period assuming that the returns to
individual characteristics remained the same as in the rst period, while allowing for the individual
characteristics to change.

6.11 Machine Learning and Big Data

Use of machine learning methods has begun to appear in the regional economics literature. One of the
earliest applications involved the use of regression trees to identify convergence clubs (Durlauf et al.,
2005). That paper is important for signaling a change in the receptiveness of economists to the use of
data instigated approaches to economic analysis. Previously dismissed as measurement without theory ,
these new machine learning methods are increasingly seen as being novel approaches to uncover
empirical patterns that have been previously unrecognized. These new patterns, in turn, provide the
basis for more exible process-based modeling of regional economic growth, and the fusion of machine
learning approaches with econometric models o ers new avenues to address questions surrounding
model stability and process heterogeneity.
Alongside the adoption of machine learning methods, has been a recognition of the potential for new
types of data sources to advance the study of regional economic development. Drawing on the big data
revolution, economists have begun to tap into new forms of data that are now becoming available at
spatial extents, scales, and temporal frequencies that were previously beyond reach. A prominent
example is the use of remote sensing imagery and pattern recognition algorithms to provide scalable
analyses of regional growth (Lessmann and Seidel, 2017). The question of the inverted-U relationship
between the level of spatial inequality in a country and the country’s level of economic development is
revisited through an expanded cross-section of regions made possible using remote sensing data on
night-time lights. The approach is to rst estimate a model relating regional incomes to luminosity for
countries with established regional accounts. This relationship is then employed to generate estimates
for income in regions in developing countries where regional accounts are unavailable or of low quality,
yet remote sensing data are available. This substantially expands the empirical basis for the inequality-
development analysis since the traditional sample bias that had limited the analysis to more developed
economies with o cial regional accounts is now removed.

7 Conclusion
As policy makers move forward in their e orts to improve living conditions and economic growth across
regions of a country, acknowledging and understating that the mechanisms through which regional
economic growth can be achieved di er from those that lead to national growth. Researchers and policy
analyses that seek to such policies must consider that assumptions of traditional neoclassical growth

31
models pose serious constraints for the analysis of regional economies. For regions, interregional factor
endowments, and di erences in levels of factor productivity, may account for the advantage or
constraints faced by an area compared to the rest of the country, or the world.
We have provided here a motivation as to why di erent methods are needed to understand the
challenges of regions. It then follows to review some of the methods available to understand and assess
the challenges regions face. While this review is not comprehensive, it provides the reader a set of tools
that can be used to better understand the works of regional economies. These tools can be used to
better asses needs of regions and carefully determine the possible impact of policies. We have identi ed
several avenues to improve our understanding of regional economies. First, a better understanding of
the economic structure of local economies is key for policy design. The layout of the interindustry and
inter-regional industrial structure and trade relationships play a critical role in determining the e ects of
various policies and programs. Second, a complete description of regional economic structure begins
with an inventory of employment, compensation (wages and salaries), and output by industry. This will
deepen the understanding of how specialized or diversi ed is the regional economy, what are its
comparative and absolute advantages, it will facilitate the development of behavioral models for
assessing impacts of policies, and it will increase the understanding of inter-regional linkages. Regional
accounting systems appear as a fourth tool that can contribute to the accurate identi cation of
interactions that take place both within the region and those that cross the regions borders in the form
of imports and exports. Finally, the spatial dimension of regional analysis can leverage the use of
geospatial information and bring additional information to discussions such as inequality or con ict that
often are engrained in spatially con ned areas but for which the spatial dimension is often ignored in
economic analysis.
While most of these methods have been widely applied in the developed world, more work is needed in
developing countries to understand how regional economies work and how methodologies applicable in
developed countries fare when used to analyze developing countries’ challenges. A lack of data has long
been an important constraint in the developing world, but new data sources promise to address this gap.

32
References
Anselin, L. (1995). Local indicators of spatial association-lisa. Geographical Analysis, 27(2):93 115.

Archibald, G. (1972). On regional economic policy in the United Kingdom. Essays in Honour of Lord
Robbins, pages 224 245.

Archibald, G. C. (1969). The phillips curve and the distribution of unemployment. The American
Economic Review, 59(2):124 134.

Armstrong, H., Taylor, J., and Taylor, J. (2000). Regional economics and policy.

Atkinson, A. B. (2015). Inequality. Harvard University Press.

Bar , R. A. and Knight, III, P. L. (1988). Dynamic shift-share analysis. Growth and change, 19(2):1 10.

Basile, R. and Lim, J. (2017). Nonlinearities in interregional migration behavior: evidence from the united
states. International Regional Science Review, 40(6):563 589.

Bekele, G. W. and Jackson, R. (2006). Theoretical perspectives on industry clusters.

Beveridge, W. (1944). Full Employment in a Free Society. G. Allen, London.

Borts, G. H. and Stein, J. L. (1964). Economic growth in a free market.

Bosker, M. (2007). Growth, agglomeration and convergence: A space-time analysis for European regions.
Spatial Economic Analysis, 2(1):91 100.

Bosker, M. and Krugell, W. (2008). Regional income evolution in South Africa after apartheid. Journal of
Regional Science, 48(3):493 523.

Brakman, S., Garretsen, H., and Van Marrewijk, C. (2010). References across the fence: measuring the
dialogue between economists and geographers. Journal of Economic Geography, 11(2):371 385.

Capello, R. (2015). Regional Economics. Routledge.

Carrillo, P. E. and Rothbaum, J. L. (2016). Counterfactual spatial distributions. Journal of Regional Science,
56(5):868 894.

Carrington, A. (2006). Regional convergence in the European Union: A stochastic dominance approach.
International Regional Science Review, 29(1):64 80.

Chandra, S. (2003). Regional economy size and the growth instability frontier: Evidence from Europe.
Journal of Regional Science, 43(1):95 122.

Chiang, S.-h. (2009). The e ects of industrial diversi cation on regional unemployment in Taiwan: is the
portfolio theory applicable? The Annals of Regional Science, 43(4):947.

Chinitz, B. (1961). Contrasts in agglomeration: New york and pittsburgh. The American Economic Review,
pages 279 289.

Conroy, M. E. (1974). Alternative strategies for regional industrial diversi cation. Journal of Regional
Science, 14(1):31 46.

33
Czamanski, S. (1971). Some empirical evidence of the strengths of linkages between groups of related
industries in urban-regional complexes. In Papers of the Regional Science Association, volume 27,
pages 136 150. Springer.

Czamanski, S. and Ablas, L. A. (1979). Identi cation of industrial clusters and complexes: a comparison of
methods and ndings. Urban studies, 16(1):61 80.

Czamanski, S. and Czamanski, D. Z. (1976). Study of spatial industrial complexes, volume 2. Halifax, NS:
Institute of Public A airs, Dalhousie University.

Czamanski, S., Czamanski, D. Z., and Ellis, S. B. (1974). Spatial Organization of Industries. I. Study of
Clustering of Industries. Halifax, N.S. : Institute of Public A airs, Dalhousie University.

Deuskar, C., Baker, J. L., and Mason, D. (2015). East Asia’s changing urban landscape: Measuring a decade
of spatial growth. World Bank Publications.

Diamond, J. (1974). The analysis of structural constraints in developing economies: a case study. Oxford
Bulletin of Economics and Statistics, 36(2):95 108.

Duranton, G. and Venables, A. J. (2018). Place-based policies for development. The World Bank.

Durlauf, S. N., Johnson, P. A., and Temple, J. R. (2005). Growth econometrics. Handbook of economic
growth, 1:555 677.

Egger, P. and Pfa ermayr, M. (2006). Spatial convergence. Papers in regional science, 85(2):199 215.

Esteban, J.-M. and Ray, D. (1994). On the measurement of polarization. Econometrica: Journal of the
Econometric Society, pages 819 851.

Ezcurra, R. (2009). Polarization trends across the us states. Applied Economics Letters, 16(9):897 902.

Feser, E. J. and Bergman, E. M. (2000). National industry cluster templates: A framework for applied
regional cluster analysis. Regional studies, 34(1):1 19.

Fields, G. S. (2010). Does income mobility equalize longer-term incomes? new measures of an old
concept. The Journal of Economic Inequality, 8(4):409 427.

Fingleton, B. (1999). Estimates of time to economic convergence: an analysis of regions of the European
Union. International Regional Science Review, 22(1):5 34.

Fingleton, B. and Palombi, S. (2013). Spatial panel data estimation, counterfactual predictions, and local
economic resilience among british towns in the Victorian era. Regional Science and Urban Economics,
43(4):649 660.

Florida, R. (2005a). Cities and the creative class. Routledge.

Florida, R. (2005b). The world is spiky. Globalization has changed the economic playing eld, but hasn’t
leveled it. Atlantic monthly, 296(3):48.

Frick, J. R. and Goebel, J. (2008). Regional income strati cation in uni ed germany using a gini
decomposition approach. Regional Studies, 42(4):555 577.
Friedman, T. L. (2005). The world is at: A brief history of the twenty- rst century. Macmillan.

34
Fujita, M., Krugman, P. R., and Venables, A. J. (2001). The spatial economy: Cities, regions, and
international trade. MIT press.

Goodchild, M. F., Anselin, L., Appelbaum, R. P., and Harthorn, B. H. (2000). Toward spatially integrated
social science. International Regional Science Review, 23(2):139 159.

Hammond, G. W. and Thompson, E. (2002). Mobility and modality trends in us state personal income.
Regional studies, 36(4):375 387.

Higgins, B. (1973). Trade-o curves and regional gaps. na.

Hirschman, A. O. (1958). The strategy of economic development. Technical report.

Jackson, R. (2015). Are industry clusters and diversity strange bedfellows?

Jackson, R. and Haynes, K. (2009). Shift-share analysis. In Kitchin, R. and Thrift, N., editors, International
Encyclopedia of Human Geography, pages 125 131. Elsevier, Oxford.

Jackson, R. W. (1984). An evaluation of alternative measures of regional industrial diversi cation. Regional
Studies, 18(2):103 112.

Kanbur, R., Rhee, C., and Zhuang, J. (2014). Inequality in Asia and the Paci c: Trends, drivers, and policy
implications. Routledge.

Krugman, P. (1991). Geography and trade. MIT Press, Cambridge, MA.

Lahr, M. L. and Dietzenbacher, E. (2017). Structural decomposition and shift-share analyses: Let the
parallels converge. In Regional Research Frontiers-Vol. 2, pages 209 220. Springer.

Le Gallo, J. and Ertur, C. (2003). Exploratory spatial data analysis of the distribution of regional per capita
GDP in Europe, 1980 1995. Papers in regional science, 82(2):175 201.

Lessmann, C. and Seidel, A. (2017). Regional inequality, convergence, and its determinants a view from
outer space. European Economic Review, 92:110 132.

Maasoumi, E. (1998). On mobility. Statistics Textbooks and Monographs, 155:119 176.

Malizia, E. E. and Feser, E. J. (1999). Understanding local economic development.

Martin, R. and Sunley, P. (2003). Deconstructing clusters: chaotic concept or policy panacea? Journal of
economic geography, 3(1):5 35.

McCann, P. (2015). The regional and urban policy of the European Union: Cohesion, resultsorientation
and smart specialisation. Edward Elgar Publishing.

Mera, K. (1967). Tradeo between aggregate e ciency and interregional equity: A static analysis. The
Quarterly Journal of Economics, 81(4):658 674.

Mera, K. (1973). Trade-o between aggregate e ciency and interregional equity: The case of Japan.
Regional and Urban Economics, 3(3):273 300.

Moore, B. and Rhodes, J. (1974). The e ects of regional economic policy in the United Kingdom.
Regional Policy and Planning for Europe. Saxon House, Farnborough.

35
Moser, M. and Schnetzer, M. (2017). The income inequality nexus in a developed country: smallscale
regional evidence from Austria. Regional Studies, 51(3):454 466.

Mossi, M. B., Aroca, P., FernÆndez, I. J., and Azzoni, C. R. (2003). Growth dynamics and space in brazil.
International Regional Science Review, 26(3):393 418.

Nazara, S. and Hewings, G. J. (2004). Spatial structure and taxonomy of decomposition in shiftshare
analysis. Growth and change, 35(4):476 490.

Perroux, F. (1955). Note sur la notion de" p le de croissance". Øditeur inconnu.

Piketty, T. (2017). Capital in the Twenty-First Century. The Belknap Press of Harvard University Press.

PolŁse, M. and Shearmur, R. (2006a). Growth and location of economic activity: The spatial dynamics of
industries in Canada 1971 2001. Growth and Change, 37(3):362 395.

PolŁse, M. and Shearmur, R. (2006b). Why some regions will decline: A Canadian case study with
thoughts on local development strategies. Papers in Regional Science, 85(1):23 46.

Porter, M. (1990). e.(1990) the competitive advantage of nations. Harvard Business Review,
68(2):73 93.

Quah, D. et al. (1993). Galton’s fallacy and tests of the convergence hypothesis (now published in the
Scandinavian journal of economics 95 (4), 1993, pp. 427-443.). Technical report, Suntory and Toyota
International Centres for Economics and Related Disciplines.

Quah, D. T. (1996). Aggregate and regional disaggregate uctuations. Empirical Economics,

21(1):137 159.

Rasmussen, P. N. (1956). Studies in inter-sectoral relations, volume 15. North-Holland.

Rey, S. J. (2001). Spatial empirics for economic growth and convergence. Geographical analysis, 33(3):195
214.

Rey, S. J. (2014). Fast algorithms for a space-time concordance measure. Computational Statistics, 29(3-
4):799 811.

Rey, S. J. (2016). Space time patterns of rank concordance: Local indicators of mobility association with
application to spatial income inequality dynamics. Annals of the American Association of Geographers,
106(4):788 803.

Rey, S. J. (2018). Bells in space: The spatial dynamics of us interpersonal and interregional income
inequality. International Regional Science Review, 41(2):152 182.

Rey, S. J. and Dev, B. (2006). σ-convergence in the presence of spatial e ects. Papers in Regional Science,
85(2):217 234.

Rey, S. J. and Janikas, M. V. (2006). Stars: space time analysis of regional systems. Geographical analysis,
38(1):67 86.

Rey, S. J., Kang, W., and Wolf, L. (2016). The properties of tests for spatial e ects in discrete markov chain
models of regional income distribution dynamics. Journal of Geographical Systems, 18(4):377 398.

36
Rey, S. J. and Le Gallo, J. (2009). Spatial analysis of economic convergence. In Palgrave handbook of
econometrics, pages 1251 1290. Springer.

Rey, S. J., Murray, A. T., and Anselin, L. (2011). Visualizing regional income distribution dynamics. Letters
in Spatial and Resource Sciences, 4(1):81 90.

Rey, S. J. and Smith, R. J. (2013). A spatial decomposition of the gini coe cient. Letters in Spatial and
Resource Sciences, 6(2):55 70.

Richardson, H. W. (1978). Growth centers, rural development and national urban policy: a defense.
International Regional Science Review, 3(2):133 152.

Rodr guez-Pose, A. and Wilkie, C. (2018). Strategies of gain and strategies of waste: What determines the
success of development intervention? Progress in Planning.
Scho eld, J. A. (1989). Federal regional development policies in Canada: Employment impact in Quebec
and the Atlantic Provinces. Review of Urban & Regional Development Studies, 1(1):65 79.

Shorrocks, A. and Wan, G. (2005). Spatial decomposition of inequality. Journal of Economic

Geography, 5(1):59 81.

St. Louis, L. V. S. (1980). A measure of regional diversi cation and e ciency. The Annals of Regional
Science, 14(1):21 30.

Stiglitz, J. E. (2012). The price of inequality: How today’s divided society endangers our future. WW
Norton & Company.

Temurshoev, U. and Oosterhaven, J. (2014). Analytical and empirical comparison of policy-relevant key
sector measures. Spatial Economic Analysis, 9(3):284 308.
Thirlwall, A. (1969). Demand disequilibrium in the labour market and wage rate in ation in the united
kingdom (1). Bulletin of Economic Research, 21(1):66 76.

Thirlwall, A. (1970). Regional phillips curves 1. Bulletin of the Oxford University Institute of Economics &
Statistics, 32(1):19 32.

Thirsk, W. R. (1973). Regional Dimensions of In ation and Unemployment: A Research Report Prepared
for the Prices and Incomes Commission. Information Canada.

Tobler, W. R. (1970). A computer movie simulating urban growth in the Detroit region. Economic
geography, 46(sup1):234 240.

Vanderkamp, J. (1971). Migration ows, their determinants and the e ects of return migration. Journal of
Political Economy, 79(5):1012 1031.

Villaverde, J. and Maza, A. (2012). Chinese per capita income distribution, 1992 2007: A regional
perspective. Asian Economic Journal, 26(4):313 331.

Woodward, D. P. (2012). Industry location, economic development incentives, and clusters. Review of
Regional Studies, 42(1):5 23.

Woodward, R. S. (1973). E ective location subsidies: An evaluation of DREE industrial incentives.

37
Woodward, R. S. (1974). The capital bias of DREE incentives. Canadian Journal of Economics, pages 161
173.
Woodward, R. S. (1975). The e ectiveness of DREE’s new location subsidies. Canadian Public
Policy/Analyse de Politiques, pages 217 230.

World Bank (2014). Urban China: Toward e cient, inclusive, and sustainable urbanization. World
Bank Publications.

Young, A. T., Higgins, M. J., and Levy, D. (2008). Sigma convergence versus beta convergence: Evidence
from us county-level data. Journal of Money, Credit and Banking, 40(5):1083 1093.

Yu, J. and Jackson, R. (2011). Regional innovation clusters: A critical review. Growth and Change,
42(2):111 124.

38
Annex

IO-Snap Software
The organization of the IO-Snap menu structure is presented below as a description of IO-Snap
functionality. Starred (∗) items open additional user input forms.
File
Load Data∗

Save Data∗

Excel∗ under development

Print∗

Preferences∗

Program Reset
Exit
Data
IO Accounts∗ State

Industry Data
*
By Region∗
*
Compensation Rates (All Regions)
*
Employment (All Regions)
*
Gross Industrial Product (All Regions)
*
FTE-Job Ratios
Gross Domestic Product
Requirements Tables∗

Multipliers∗

Sector Distributions
*
Industry Accounts
*
Commodity Accounts
*
Industry Labor
*
Commodity Trade Balance
Edit

39
 *
Aggregation∗
*
Modify Data

• Use∗

• Make∗
• Final Demand∗
Analyze
Regionalize∗

Impacts∗

*
Type 1 Industry-Driven∗
*
Type 2 Industry-Driven∗
*
Type 1 Commodity-Driven∗
*
Type 2 Commodity-Driven∗
Help
About
Documentation
Activate∗

Deactivate∗

REAL-IO Software
The available REAL-IO operations and analyses are organized as follows:
Table operations
Displaying I-O tables
Displaying industrial structure of target regions by value-added and Output
Single region (country) analysis
Import penetration ratio
Key sector analysis
Leontief inverse (using purchase coe cients)/ Backward linkages
Goshian inverse (using sales coe cient)/ Forward linkages
Import content shares of exports
Estimates the leakage through import when export activity increase

40
 Induced value-added by exports
Estimates the additional value added generated by increased export activities
Labor multipliers
Estimate the indirect and induced e ects of direct changes in labor by sector
Field of in uence
Estimates the economy-wide impact of change in individual, multiple, row or column coe
cients
Average propagation link
Estimates the rounds of spending generated by an expansion in each sector
RAS procedure to update tables
Provides a method to update input-output tables given information on the margins for a
future year
Multi region (country) analysis
Interregional spillover e ects
Estimates the spillover e ects from changes in one region or sector
Average propagation link of multiregional framework
Estimates the rounds of spending generated by expansion in one sector in one region
Under Development
Feedback loop analysis
Decomposes the structure of interdependence between sectors into a hierarchical system of
loops
Structural decomposition analysis
Decomposes change in an economy into contributions by technology demand and their
interactions
Hypothetical extraction
Explores the impacts of the removal of a sector (single economy) or a region (multiregional
system) on macro measures (output, income, employment)

41