Making the Most of Coresident Data: Credible Evidence on Intergenerational Mobility with Sibling Correlation

Abstract

Many high quality data sets such as census and Demographic and Health Survey (DHS) are usually not used for estimating intergenerational mobility in developing countries owing to concerns about sample truncation bias in coresident data. Using four exceptional data sets that include nonresident children, we report the first evidence that the bias in estimated sibling correlation, a broad measure of relative mobility, is small in coresident samples (3.16%), smaller than that in intergenerational regression coefficient (10.51%) and intergenerational correlation (4.28%). We offer an explanation for this finding: sample truncation causes downward bias in both the numerator and denominator of the sibling correlation formula, largely canceling each other out. We also find that the bias in sibling correlation is smaller for the younger cohorts, but this does not hold for two widely-used measures of relative mobility: intergenerational regression coefficient and intergenerational correlation. Cross-country mobility ranking based on sibling correlation estimates from coresident samples is reliable, giving the correct ranking in 91 percent times. Our analysis has far-reaching implications for research on intergenerational educational mobility as sibling correlation is used by both economists and sociologists as an omnibus measure of mobility.

Keywords: Coresident Data, Sample Truncation Bias, Intergenerational Educational Mobility, Sibling Correlation, IGRC, IGC, Developing Countries, Census, DHS, LSMS

JEL Classification: J62, O12