When measure matters: coresident sample selection bias in estimating intergenerational mobility in developing countries

Potential biases from coresident sample selection have been a major stumbling block for research on intergenerational mobility in developing countries. We use two rich data sets from Bangladesh and India to provide evidence on the extent of coresidency bias in standard measures of intergenerational mobility: intergenerational regression coefficient (IGRC) and intergenerational correlation (IGC). Estimates for all children, father-son, and mother-daughter persistence in schooling show that the IGRC estimates are severely biased downward (average 30 percent). In contrast, the bias in IGC estimates is much lower (average less than 10 percent, in many cases less than 5 percent). Truncation due to coresidency criterion in a survey biases the IGRC estimate downward, but it also biases upward the estimate of the ratio of the standard deviations of parental to children's schooling. The IGC estimate suffers from lower bias because the upward bias in the estimate of the ratio of standard deviations partly cancels out the downward bias in the IGRC estimate. The evidence suggests that the available household surveys in developing countries can be fruitfully used to understand intergenerational mobility if one focuses on IGC. The findings have important implications for cross-country comparison of intergenerational economic mobility.