Inferring Geographic Interdependencies for Retailing
David Mushinski Department of Economics, Colorado State University
Stephan Weiler Department of Economics, Colorado State University
Benjamin Widner Department of Economics, New Mexico State University
Understanding the dynamics of regional trade is important for designing development programs. Identifying “export” industries that sell to other regions can help to determine where to target dollars. Understanding trade patterns, for example, can help in predicting sales tax revenues. Different firms have different market areas, based on the product/service transport cost intensity, fixed costs of production and distribution, and topographical and infrastructural features of the region. These varying market areas lead to a distribution of businesses where high-order locations include the broadest variety of firms and products, and are the places in which high fixed cost, widely distributed good- and service-producing firms choose to locate. Low-order locations only include those products and services with minimal fixed costs and a highly localized service area.
Retail establishments are of special interest in this regard, given the preferences of consumers for convenience revealed by the decision to shop locally. Retail activity is also of policy interest, due to the importance of sales taxes to public finance. Retailing shows sectoral interdependence in the form of multi-purpose shopping, where a number of retailers are located in close proximity in, say, a shopping center, often located in a larger town. Such an arrangement improves efficiency by reducing transportation costs, as consumers can combine shopping into one trip. Alternatively, retailing may be exhibit low geographic concentration for frequently purchased products bought between major shopping trips.
Empirical identification of the relationship between an industry and surrounding market areas depends on the relevant market area of that industry. Until recently, data on an industry’s possible market area has been imprecise due to data limitations. For example, neighboring areas are often defined as the remainder of a county in which they are located. But one would not expect this definition to be accurate except by sheer coincident. Fortunately, the good news is that the recent development of Geographic Information System (GIS) tools makes it possible to extract data for market areas of varying radii, so that we can now better approximate a retailer’s market area.
Our analysis included data for two different geographic units. The first unit concerned the places on which our analysis focused. We focused on places with 2500 people or more because Economic Census data were available only for places of those sizes. In addition, this minimum size also usefully narrows the focus to such places that have at least some probability of appearing on the Central Place hierarchy for our focal industries. We also generally sought non-metropolitan places (including their neighboring areas) which were relatively independent economic units in the sense that their economies were spatially separated from other places in our sample, while still retaining a proportion of overlapping areas to provide a full spectrum of potential interactions. In our final sample, 73% of focal places were at least 30 driving minutes apart, with no such places overlapping across state lines.
The second geographic unit in our analysis was the neighboring areas around the original location. Our analysis included two regresses from neighboring areas: (1) the number of establishments in neighboring areas (ESTN), obtained from the Economic Census, and (2) the population in neighboring areas (POPN), obtained from the Decennial Census, The neighboring areas were defined in terms of specific travel-time radii around the place. The radii were calculated using all navigable roads leading away from the place. The radii were determined in the discrete one-way ten minute increments (20, 30, 40, 50, 60, 80, and 120 minutes of one-way travel) using Microsoft’s MapPoint program.
The Neighbor Variables provide insights into geographic interdependencies and the extent to which a retail industry produces for export. We would expect ESTN to be negatively related because an increase in the number of establishments in neighboring areas in an industry would be expected to reduce the need for people in neighboring areas to travel. The second Neighbor Variable was the population in those areas (POPN). This variable captures the impact of an increase in the number of potential customers in a neighboring location.
We estimated a model for an industry for market areas of differing radii and then use statistical techniques to identify the appropriate market radius of that industry. Our test provided insight into the geographic market interdependencies between places and their neighboring areas. If the neighboring market area is specified too narrowly, then little evidence of market interdependencies is present. For broader market areas, some interdependencies would be missed if one did not test for an industry’s optimal market area. Our results indicate that many of the retail industries studied produce both for local consumption and for export to other regions. In this regard, we distinguish between weak hybrid retail industries and strong hybrid retail industries. The former are weak in the sense that they produce for export until establishments like them appear in neighboring areas, at which point their export component dissipates. Strong hybrid retail industries are stronger in the sense that they will produce for export even if competing establishments appear in neighboring areas.