Self-selection, migration and inequality in source regions

Abstract

This dissertation provides theoretical and empirical explorations of the impacts that the internal migration has on inequality in source regions. Chapter 2 identifies the lack of microeconomic contents, especially those on intra-household economic linkages as the primary reason why the Roy model may give misleading predictions on the impacts in developing countries like China. Keeping this drawback in mind, Chapters 3, 4 and 5 extend the Roy model by combining its statistical assumptions with simple yet stylized models capturing decisions of income-maximizing rural households constrained by factor endowments. Comparing to the original model, the extended models have more solid microfoundations, they relate also to a different statistical problem, i.e. censoring rather than truncation. The difference in statistical structure leads to different predictions on the impacts. Particularly, in an empirically relevant setting where different types of labor inputs are heterogeneous in productivity but perfectly substitutive in household agriculture, Chapter 4 shows that unlike the original model, the extended model admits decreasing, constant and increasing rural inequality. Chapter 4 highlights also the non-causal relationships between the pattern of selection and the direction of change of rural inequality. Chapter 6 tests these relationships with the Chinese Health and Nutrition Survey (CHNS) data. It is found that the pattern of selection of rural-to-urban migration in China could undergo two transitions between 1991 and 2009. Nevertheless, the directions of change of rural inequality predicted using patterns of migration selection could disagree with the data in certain sub-period. Further analysis attributes much of these disagreements to a well-known puzzle that the actual size of migration is often smaller than its prediction based on observed intersectoral wage gap. Chapter 6 offers thus several preliminary explanations to help resolve this puzzle.