Abstract
This article uses a discrete-time multivariate duration model to study poverty transition in rural China between 1989 and 2006. The analysis identifies nonlinear negative duration-dependence for both exit and re-entry rates of poverty. There is significant difference in hazard rates of exit and re-entry associated with geographic location and educational level of households, but less related to gender, occupation or ethnic background of household head. The factors facilitating households' ending a poverty spell are found to be education, land ownership, asset accumulation, health insurance and outmigration, whereas larger family size and dependence ratio may reduce the chance of exit.
Notes
1As Bigsten and Shimeles (Citation2008), there is presumably no correlation between repeated spells for the same household over time, that is, independence between multiple spells. In fact, we split households into subjects with single-spells and then pooled them for estimation.
2The concept employed here is in line with Devicienti (Citation2002, Citation2010) and Bigsten and Shimeles (Citation2008). A household just starting a (non)poverty spell at t means it was in (non)poverty at t – 1 and shifts out of this state at t. Our sample contains seven waves of the surveys. Therefore, the first (non)poverty spell starts at the second wave and the maximum duration is 5.
3Households' initial (non)poverty status is assumed to be exogenous to their characteristics. Devicienti's (Citation2010) model controls for endogeneity of initial conditions, which may lead to our future research.
4We also experimented with Gamma and Heckman and Singer's (Citation1984) mixed mass-point distributions but maximization procedures failed to converge to a solution.
5Coastal provinces are Jiangsu and Shandong. Central provinces are Henan, Hubei and Hunan. Western provinces are Guangxi and Guizhou.