Measurement error in income and consumption data and economic mobility.

摘要

A sizable volume of literature on the studies of poverty dynamics and economic mobility has developed through the application of panel data surveys. Very few researchers, however, have provided solutions to the measurement error bias generated by surveyed income and consumption, although the presence of such bias has been widely acknowledged. My dissertation uses data from the Korean Labor and Income Panel Study (KLIPS) to examine whether measurement error in income and consumption data has the potential to generate biases for studies on poverty dynamics and economic mobility.;A first-differenced dynamic panel model is estimated, and it suggests that there is substantial time-varying measurement error in reported income and consumption, leading to a bias towards zero in the estimates of income and consumption dynamics. The variance of time-varying measurement error is as large as the variance of the equation error for both income and consumption dynamics. This result also supports the view that time-varying measurement error exists in reported income and consumption and has a substantial magnitude. Interestingly, the standard deviation of time-varying measurement error, as well as that of the equation error, is much larger in the model for the income dynamics than that for consumption dynamics. This result suggests that time-varying measurement error is more prevalent and varies more across households in income than in consumption and also indicates that households smooth their consumption relative to their income in the face of shocks.;Finally, I also investigate potential measurement error biases in poverty transition matrices. Transition matrices based on survey consumption are compared to ones based on measurement-error-free simulated consumption which is built on initial conditions and parameters estimated from a basic consumption dynamics model allowing for measurement error. The study finds that measurement error in consumption data magnifies economic mobility in and out of poverty. Roughly 44% of households initially in poverty at time t-1 are found to be out of poverty at time t using consumption data. However, when measurement error is removed through a model-based simulation, only between 16 and 20% of households initially in poverty are found to be out of poverty.