|
POSITION STATEMENT
As an urban economist and applied econometrician, nearly all of my research has a spatial dimension. One focus of my research is the effect of zoning on the spatial pattern of land values. Another focus is the effect of employment subcenters on the spatial patterns of population and employment densities in polycentric cities. In a recent project, I analyze how housing prices changed when a copper smelter closed and later when it was designated as a superfund site. My research has also been focused on analytical tools for spatial models, including parametric approaches to discrete choice spatial models and applications of nonparametric approaches to the analysis of spatial data.
Taking space into account is critical when analyzing externalities. For example, in our analysis of zoning in suburban Chicago, John McDonald and I found that areas were more likely to be zoned for manufacturing use at the borders between suburbs (Journal of Urban Economics, 1991). Some of the pollution, noise, and traffic generated by manufacturing plants can be pawned off on a suburb's neighbors. We have found that highly disaggregated data are desirable when analyzing zoning. Many studies fail to distinguish between areas within a suburb by treating the municipality as the unit of analysis. In fact, some sites within suburbs are best suited for a particular use, and we have found that these sites are likely to be zoned for the use that produces the highest land value.
My work with Paul Thorsnes is another example of the importance of taking space into account when modeling the effects of externalities (Advances in Econometrics, 2000). In a recent manuscript, we find that the spatial pattern of housing prices in Tacoma, Washington were affected significantly by distance from a copper smelter. The smelter spread arsenic over nearby ground and the emissions reeked. We find a positive gradient for distance from the smelter up to the closing of the smelter, after which the gradient falls to insignificance. Superfund site designation and the start of cleanup operations had no effect on the gradient.
These two examples are representative of the situations where spatial modeling adds important insight to an analysis of externalities. Externalities nearly always have a spatial dimension. Local public goods such as parks, schools, neighborhood pools, snow removal, public hospitals, mosquito abatement, and municipal golf courses all generate benefits that decline with distance. Similarly, the negative externalities associated with crime, pollution, and traffic congestion are geographically concentrated. Both positive and negative externalities affect the behavior of local government, firms, and individuals.
Though significant advances have been made in approaches to spatial data analysis, important problems remain. First, work still remains in developing models and estimation procedures that are appropriate for large data sets. Current spatial approaches generally involve ad hoc specifications developed to analyze small data sets, and often are quite cumbersome when applied to large data sets. This feature of current models is ironic because the most common approaches use maximum likelihood techniques to replace simple ordinary least squares estimation, and these techniques are grounded in asymptotic theory.
A second problem is the difficult in distinguishing between common spatial models and general model misspecification. Spatial autocorrelation may well be the result of a misspecified functional form. Explanatory variables in spatial models are typically highly correlated across space, making results very sensitive to model specification. As an example, consider the excellent study by Brueckner (Journal of Urban Economics, 1998), who analyzes strategic interactions between neighboring municipalities in the adoption of growth controls. Many variables may explain the stringency of growth controls, only some of which are available to the researcher. Missing variables are likely to be correlated across space. Is an indication of strategic interaction real or merely the result of missing variables or functional form misspecification?
Nonparametric procedures are currently underdeveloped in spatial analysis. These procedures impose less structure than common spatial models and are fairly easy to apply to large data sets. Nonparametric statistics is still largely theoretical and difficult for the average practitioner to read. Suitable software is not publicly available. CSISS could play a pivotal role in making nonparametric procedures accessible to applied researchers through dissemination of lecture notes, working papers, and computer code. A similar role would be helpful in making other approaches - such as GMM estimation - accessible to applied researchers. In general, spatial analysis is hampered by the lack of suitable computer code. SpaceStat is an excellent program, but many researchers prefer to use one computer program for all applications. It would be helpful for CSISS to serve as a warehouse for computer code for implementing common spatial procedures.
|