A Study of Spatial-Autocorrelation Model Applied To Vegetable Planting-Area Variations in Okinawa-Island

Abstract

Spatial Autocorrelation Analysis is a method which takes relation structures between objects into conventional determinant-finding researches. The important point of this analysis is to specify efficiently the appropriate and representative spatial-autocorrelation pattern as weight matrix. For the purpose of establishing this model as practical method, we applied it to explain variations in planting area of green beans and bitter gourd galsam pears in 33 municipalities of Okinawa Island. Since Okinawa is relatively isolated and can be regarded as a closed system, bias from other prefectures in the mainland can be negligible. Green beans regarded as the most strategic vegetables for mainland metropolitan markets, while bitter gourd galsam pears, which were traditional Okinawan domestic vegetable, rapidly developed to the second largest strategic vegetable after deregulation of epidemic prevention in 1990. Therefore we can expect high interdependency in planting area variations among the municipalities due to their competition or cooperation.
From practical viewpoint, we tried to test various patterns of spatial-autocorrelation-matrices based on gravity type and adjacency type. Some of them showed statistical significance and, totally, the adjacency type matrixes showed better statistical reform performance to conventional regression model.
Furthermore, the statistical difference of relation matrices and spatial autocorrelation between two vegetables seemed to reflect the real vegetable diffusion process.
Finally, we suggest some topics of plausible relation matrix preparations and the case of panel data treatment.