| Back

Modeling Interregional Migration in China

Jianfa Shen

Migration is an important component of population change and has close relations with urban and regional development. Internal migration within a country is also a complicated phenomenon involving migration flows among various cities and regions. Many efforts have been made to model migration flows using distance, population, and other factors at origin and destination to explain migration. Much methodological advancement has been made in the modeling and analysis of interregional migration. However, previous migration modeling has been done in a black-box. The overall performance of a migration model is evaluated with the contribution of all explanatory variables including regional attributes and spatial interaction effect. No detailed research has been made to examine the following. What are the contributions of spatial interaction and regional attributes to migration? How good have the regional attractiveness, regional emissiveness and spatial interaction been modeled? Which part has been modeled much better?

Professor Jianfa Shen, Professor in GRM Department and Director of Research Centre for Urban and Regional Development, Hong Kong Institute of Asia-Pacific Studies, has completed a research on Modeling Interregional Migration in China to fill above research gap. The project titled "Modeling Interregional Migration in China 2000-2005: Analyzing the Modeling Error of Regional Attributes and Spatial Interaction" was funded by Research Grants Council of Hong Kong SAR (RGC Ref. No. CUHK451912) for the period 1st January 2013 to 30th June 2015.

This research developed a new method to estimate the migration modeling errors by their sources. The research overcame the problem of spatial autocorrelation in migration modeling and used the notion of migration spatial structure to estimate the effect of spatial interaction. Following the notion of migration spatial structure, the observed or estimated regional migration matrices of a migration system can be fully described by four main factors: the overall effect, the relative emissiveness and the relative attractiveness of specific regions, and the effect of spatial interaction between pairs of regions. By calculating the contributions of migration factors to the modeling error, this research reveals which factors of the migration process can be modeled more or less accurately using the recent data of regional migration in China.

There are three steps in the error analysis. The first step is the estimation of a migration model for a specific regional migration system. Two sets of migration flows can be obtained, including one set of observed flows from census and another set of estimated flows based on the migration model. The second step decomposes each set of migration flows into the relative emissiveness, attractiveness, and the level of interaction between pairs of regions. In the third step, the relative emissiveness, attractiveness, and the level of interaction based on observed and estimated migration flows will be compared to calculate the modeling error (Fig. 1 and 2).

A network spatially filtered Poisson migration model is estimated for China. Error analysis shows that the modeling errors of the constant K, the relative emissiveness, and attractiveness caused the weighted absolute mean errors of 1.20 percent, 14.60 percent, and 15.57 percent in migration flows respectively. The spatial interaction caused the greatest weighted absolute mean error of 31.55 percent in migration flows. Thus the spatial interaction effect remains the most difficult to be modeled. More efforts should be made to improve the approach to model the effect of spatial interaction. The result has been published in the paper: Shen Jianfa, 2016, Error analysis of regional migration modeling, Annals of the American Association of Geographers, DOI: 10.1080/24694452.2016.1197767.