INDSCALによるイノベーションの拡散過程の空間分析

昭和初期の東海地方におけるラジオ聴取契約の事例

抄録

The purpose of this paper is to explore the potential applicability of Multi-Dimensional Scaling (MDS) to a spatial analysis of diffusion problem. Especially it aims at presenting an approach to analyze the diffusion of household innovation by using INdividual Difference SCALing (INDSCAL). As shown in the fact that there exists a spatial covariation between the opportunity to get innovation and the accessibility to information (Hanham and Brown, 1976), some of structural elements constituting a diffusion process of innovation would be spatially correlated. Factor analysis has usually been applied to such a spatial covariation problem, but it is not so much concerned with extracting the spatial structure itself. Accordingly, it is not necessarily a suitable method for the spatial diffusion study wherein both the functional and spatial perspectives are required (Semple and Brown, 1976). Moreover, existing methods devised to analyze the diffusion of household innovation, for example, the spatiotemporal logistic model or the space-time autoregression model, lay their emphases on the spatial perspective too much to meet this requirement. Now, assuming that there exists a space common to structural elements correlated spatially one another, INDSCAL seems to present an alternative method to cover both the perspectives. Its framework consists of the following procedures:
(1) To prepare the data for structural elements, which are considered as (dis)-similarity data.
(2) To recover a two-dimensional common space from these data by using INDS-CAL.
(3) To apply four-variable trend surface analysis incorporating the common space and time coordinate system to the cross-sectional data for the rate of diffusion.
Based on the analytical framework outlined above, this paper investigates the diffusion process of radio subscription in the Tokai district, mainly focusing its attention on the aspects of the spatial interaction and the occupation structure of cities. The study period is from 1929 to 1933, which corresponds to the primary stage of the diffusion. Forty-four cities are studied with more than ten thousand population and with more than two thousand households in Aichi, Gifu and Mie Prefectures as of 1930 (Fig. 2). Among those cities the radio station of Nagoya had started its service in the year of 1925. First, in order to recover a two-dimensional interaction space, the matrix of spatial interaction of cmmuting workers and students was multidimensionally scaled by using M-D-SCAL. In spite of high stress of 36%, which is not good, the configuration shows four significant city groups reflecting the sub-areal divi-sion of the three prefectures (Fig. 3). That is, the cities of the Mino area of Gifu Prefecture, the Mikawa area of Aichi Prefecture, the Ise area of Mie Prefecture and the Owari area of Aichi Prefecture form four clusters clockwise from the top right-hand quadrant in Fig. 3. Second, to get input data for M-D-SCAL to recover the occupation structure space, a dissimilarity index suggested by Johnston (1979) was calculated from the percentages by eight groups of occupational population; agriculture, fishery, mining, manufacturing industry, commerce, transportation, service and others. Fig. 4 shows a two-dimensional solution of the occupation structure space, whose stress is as good as 4%. An application of non-metric hierarchical cluster analysis revealed that the first dimension was interpreted as representing the contrast of occupation structure such as agriculture versus non-agriculture, and the second dimension as representing the dominance of manufacturing industry among non-agricultural occupation. The two-dimensional space can thus be identified as arranging the cities according to the difference of occupation structure.
Then the set of matrices of inter-city distances in two-dimensional Euclidean spaces for both was used as input for INDSCAL.