Multivariate spatial interaction models as applied to China's inter-provincial migration, 1982-1990

Abstract

By using spatial interaction models (SIMs) to estimate place-to-place migration, it usually means that we employ some known information to estimate migration flow patterns. The conventional spatial interaction modelling of migration has been questioned for its lack of explanatory power. The present study takes a new perspective that has not been attempted before; that is, multiple socioeconomic variables can be included in the conventional SIMs. Models based on this new approach are termed Multivariate Spatial Interaction Models (MSIMs). In this particular study, it involves using the two additional variables, the average total annual investment and migrant stock, together with total out-migrants of each province and a distance matrix, to estimate China's province-to-province migration flows. The fundamental idea behind this new perspective is to weigh the socioeconomic importance of each province, so that migration flows will not only be accounted for by the traditional spatial distance but also be accounted for by socioeconomic conditions of provinces. The proposed MSIMs are derived under the framework of the information minimisation principle. MSIMs are successfully calibrated utilising the 1982-87 and 1985-90 province-to-province migration data for the 28 provinces of China. The models are calibrated by iterative procedures written in FORTRAN 77. The MSIMs are further extended to estimate the origin-specific migration flows. The importance of the two additional variables is evaluated in terms of the relative contribution to the performance of the models. The original contribution of the present research can be understood to lie in the new proposed MSIMs, in the extension to modelling origin-specific flows, which have not attempted before, and in the successful empirical application of the models to the Chinese inter-provincial migration data. The empirical results illustrate that all the MSIMs produce better results than the conventional SIMs. In other words, all models with the additional variable(s) are capable of replicating migration flows with a much-improved degree of accuracy, in comparison with the conventional model. The calibration has therefore provided empirical support for the validity and utility of the multivariate approach to the spatial interaction modelling of migration. However, the results do not necessarily imply that more variables included in the model would result in a corresponding improvement in model performance. Furthermore, a comparison of performance level between the MSIMs and origin-specific MSIMs indicates that the estimation of origin-specific migration flows can further improve the degree of accuracy in replicating the observed migration. Major forces that influence China's inter-provincial migration are represented by the two additional variables [three quarters] migrant stock and total annual investment. These two variables are appropriate in that they reflect both migration policy change and economic development strategy. The empirical results also imply that selecting appropriate variables is crucial in calibrating migration flows within the proposed framework, because variable selection must be based on the specific country or areal contexts, on the one hand, and is also dependant upon the availability of data, on the other.