Economic Geography

Most inquiries, but not all, are driven 
by clear research questions
While most research projects start with a fairly clear research question, and the
methodology is also clear as to how to approach these research questions, in
many cases the research is exploratory, and the approach needs to be developed
as the project proceeds. A good example of work of this type that I was involved
with is a paper that resulted from the synthesis of various strands of data gathered
in a large NSF-funded project focused on the producer services. My co-author
David Lindahl and I realized that we could possibly classify responses of the firms
involved in this project into a taxonomy that would allow us to characterize their
development sequences, and possibly provide a test of some business strategy
literature, and Taylor and Thrift’s model of segmented industries (Ansoff 1965;
Taylor and Thrift 1983). Lindahl and I experimented with various combinations
of variables that were included in our database, recording our classification in
colored chalk on about 20 feet of blackboard, and discussing among ourselves
the positioning of individual businesses in this classification scheme. This exercise
involved 418 detailed questionnaires with a mixture of qualitative and quantita-
tive information, and ultimately yielded a test of the Ansoff model that showed
that by being adaptive firms were rewarded with growth. It also led to a classifi-
cation that documented a wide variety of adaptive behavior, ranging from firms
that were failing to those that were soaring in sales growth to those that were just
stagnant (Beyers and Lindahl 1997). This classification turned out to have some
similarities to that developed by Taylor and Thrift, but had its own distinctive
structure. This paper was not visualized when we began this research project, and
it was only after we began to study patterns of responses to multiple questions
that it occurred to us that we could develop these classifications of firm behavior.
In other cases my work has been purposefully exploratory. We have little in the
way of interregional trade data in the United States, and while we have many
models of regional economies, we do not have a rich legacy of multi-regional
models. I became interested in taking the bits and scraps of data on interregional
trade that came from regional input–output models, and tried to speculate about
possible interregional structures. I first developed a hypothetical interregional
interindustry matrix, with a specialized industry in each region, and a generic
188
William B. Beyers

---

local services sector (Beyers 1978). I am not aware of anyone else attempting this
kind of simulation, but it was an important effort to make, given how ‘open’
regional economies are – that is to say their trade relationships are typically much
stronger with other regions than internally. This model was set into a system of
equations that produced interregional income flows and interregional final demands,
such that the output and income distribution among the regions evolved over time.
I recall presenting this paper at the regional science meetings in Krakow, Poland,
and had lots of computer printout containing the tables from this speculative
modeling that attracted considerable suspicion from border guards who were sure
I was out to sell all those numbers! I followed up this model with one that had a multi-
regional demographic accounting model integrated with it, and explored the evolu-
tion of populations and economic activity over time with this system (Beyers 1980).
I further developed models of this nature after obtaining some support from
National Science Foundation (NSF) to explore spatial linkage patterns of busi-
nesses located in Washington State with their markets and sources of supply else-
where in the United States. This work found that the interaction among the
states showed a gravity-model like pattern for Washington firms with clients and
suppliers located in other parts of the United States, and tied the levels of activ-
ity in each region into estimates developed by Polenske in relation to the
Multiregional Input Output (MRIO) model (Polenske 1970). The models were
configured to have most of the interindustry multiplier effects be interregional
(rather than intraregional), in accord with the data from regional input–output
accounts (Beyers 1974). This kind of modeling is in many ways dreaming with
numbers, but the general properties of the results appeared not to be counterin-
tuitive. I also experimented with the use of drawings to illustrate possible alter-
native spatial linkage configurations, as opposed to using a gravity model in each
region (Beyers 1981). This work also involved the use of cluster analysis to decom-
pose the data in the national input–output model into broad categories of linkages.
It is unfortunate that statistical accounts in the United States have failed to repre-
sent more realistically trade relationships among regions.