ABSTRACT
I Introduction The longstanding decline in agricultural employment (Barkley 1990) has led to heightened interest in other potential sources of rural employment growth, especially in traditionally agriculture-dependent areas. Researchers have turned their attention to a host of both standard and novel prescriptions such as household amenities, human capital, new economy firms, and fiscal policy (Deller et al. 2001; Goetz and Rupasingha 2002; Huang et al. 2002; McGranahan 2002; Thompson et al. 2006). Yet, despite recognition of potential rural-urban differences (Ferguson et al. 2007), an unexplored aspect of this research is the degree of spatial heterogeneity (non-stationarity) in rural U.S. growth dynamics. Spatial heterogeneity may be expected to arise because local labor markets vary in their structure, social context, and histories (Lloyd and Shuttleworth 2005) in ways not readily captured by standard explanatory variables in global regressions. To paraphrase the old expression, “if you have seen one rural community, you have seen only one rural community.” Spatial heterogeneity in growth dynamics could render global estimates misleading in terms of local outcomes. For example, accepted findings with respect to the role of key variables in economic growth may be the result of global estimates (e.g., ordinary least squares, OLS) that mask significant local variation, even in the direction of influence. Alternatively, the standard estimates may suggest no marginal effect, while in reality the factor stimulates growth in some areas while reducing it in others, yielding an average effect of about zero. Aside from the importance of discovering the true nature of the relationships, successful local economic development policy requires knowledge of local socioeconomic processes and growth dynamics (Blank 2005; Nizalov and Loveridge 2005).1 A recent approach gaining popularity in accounting for potential geographic heterogeneity in socioeconomic processes is the geographically weighted regression (GWR) (Fotheringham et al. 2002). In contrast to the global regression approach, GWR can estimate separate coefficients, potentially for each observation (area). In estimating each region’s own regression, characteristics of the individual areas
included in the sub-sample are weighted by their spatial proximity. Spatial weighting smoothes variation in parameter estimates, revealing broad regional differences in the local marginal responses. Although still relatively uncommon, GWR is increasingly being applied in regional analysis. Recent applications include examinations of geographic heterogeneity in regional socioeconomic processes related to poverty (Benson et al. 2005; Farrow et al. 2005), commuting (Lloyd and Shuttleworth 2005), regional industrialization (Huang and Leung 2002), regional growth effects of agricultural policy in Western Europe (Bivand and Brunstad 2003), and local employment growth in Canada (Shearmur et al. 2006). In terms of potential spatial heterogeneity in rural growth dynamics, Deller et al. (2001) raise the possibility that there may be agglomerative or interactive growth effects of amenities. Spatial differences in such effects could produce heterogeneity in rural growth responses, and potentially differences in appropriate policies. For example, in contrast to U.S. results, Ferguson et al. (2007) find that amenities have relatively little influence on Canadian migration relative to economic factors, especially in rural Canada. Huang et al. (2002) discuss how human capital effects on growth are likely to vary regionally, possibly producing a “brain drain” in some regions. Although one could imagine using carefully selected interaction variables to detect these spatial variations with global approaches, this would require intricate knowledge of the specific set of interactions and adequate degrees of freedom, while specification problems such as multicollinearity could be exacerbated.2 In addition to the value of the GWR approach in terms of revealing spatial heterogeneity, the results can also inform global approaches. Region-specific results may provide a more detailed perspective on underlying relationships, allowing refinements in the global specification. Indeed, severe misspecification bias has been found in general spatial interaction modeling because of the spatial variation in local parameters, which could be missed in global approaches (Fotheringham 1984, 1986). Therefore, this chapter empirically assesses the spatial heterogeneity of nonmetropolitan county employment growth dynamics over 1990-2004. In particular, we hypothesize that there is significant spatial variation in the influence of climate/ natural amenity and human capital on employment growth. We compare global regression estimates with the variation in GWR estimates for growth-related factors. Among the findings of particular interest, statistically significant geographic variation in employment growth responses is found for amenities, college completion, and immigration. Interestingly, the influence of agriculture’s employment share on subsequent job growth does not vary spatially across the country. Some amenity variables are found to have insignificant global effects, suggesting little marginal impact in traditional analysis. Yet, the GWR approach reveals a rich pattern, showing that these variables may have locally statistically significant effects that (nationally) offset one another. Likewise, greater college attainment stimulates growth in some areas, while reducing it in others. Immigration effects also vary from negative to positive. Generally, we conclude that “one size does not fit all” in understanding the underlying growth processes and in
informing local economic development policymaking. Moreover, we believe our findings can help refine global specifications and that the geographical diversity of results can stimulate new hypotheses concerning rural growth processes. In particular, one question that arises is why the influences change so suddenly over geographic space-producing knife edges-even within what are thought to be relatively homogenous regions? In the next section, we develop a model of nonmetropolitan employment growth, including a discussion of how heterogeneity in growth processes can arise. Section 3 follows with the empirical implementation of the model. Section 4 presents and compares global regression results with those of GWR, including maps illustrating the geographic variation in employment growth responses to key variables. The final section summarizes the results and discusses their implications for rural economic development policymaking and for regional/ urban modeling, including ways to improve global methods.
