A study of the Missing Data Problem for Intergenerational Mobility using Simulations

Applied research on the association between parent and child lifetime income is relying on income data that covers only part of the life cycle which may lead to misleading estimates of the intergenerational elasticity (IGE). In this paper I study the bias of IGE estimates for different missing-data scenarios based on simulated income processes. Using an income process from the income dynamics and risks literature to generate two linked generations’ complete income histories, I use Monte Carlo methods to study the relationship between available data patterns and the bias of the IGE. I find that the traditional approach using the average of the typically available log income observations leads to IGE estimates that are around 40 percent too small. Moreover, I show that the attenuation bias is not reduced by averaging over many father income observations. Using just one income observation for each generation at the optimal age (as discussed in the paper) or using weighted instead of unweighted averages can reduce the bias. In addition, the rank-rank slope is found to be clearly less sensitive to missing data.