New Model Framework for a Principled Assessment of Selection Bias: Case Study on Labour Supply

Abstract

We propose a new model framework to investigate missing-data induced selection bias (SB). Drawing on the missing at random (MAR) and missing not at random (MNAR) taxonomy of the multiple imputation approach, we show that SB is predicated on MNAR and is empirically assessable by sensitivity analysis. Specifically, SB risk is assessed using various MNAR scenarios which are simulated by imposing systematic deviations on stochastically imputed missing data under the MAR condition. Taking the case of missing wage rates of non-working wives in household surveys as example, we find that SB is practically ignorable unless MNAR scenarios are extreme.

Keywords: selection bias, missing at random, multiple imputation, sensitivity analysis, wage, labor supply

JEL Classification: C21, C52, J20, J24