Examining estimator bias and efficiency for pseudo panel data: A Monte Carlo simulation approach

Abstract

Pseudo panel data have been increasingly applied in empirical research as an alternative approach to a longitudinal analysis when genuine panel data are unavailable. However, conventional techniques are typically used to estimate pseudo panel data models without careful consideration to some unique properties of pseudo panel data. Ignoring properties such as time-varying cohort effects, a small number of constructed cohorts, large between-group variance, and trade-offs between cohort sizes and number of cohorts potentially lead to estimation bias or inefficiency not observed in genuine panel data. This paper presents a Monte Carlo experiment with scenarios that are designed to generate, under conditions of limited observations, various data possessing pseudo panel data characteristics, and evaluates the performances of various estimators using the simulation results. The main research findings are that the large between-group variation of the exogenous variable and the variance of fixed group effects in pseudo panel data are the primary causes of estimation bias and inefficiency. Other factors including the cohort size and potential non-spherical errors have a smaller impact on the estimators’ performances. An empirical application using Sydney Household Travel Survey data is also presented to illustrate the simulation findings.