Sample Selection Bias Due to Differential Mortality: A Supplementary Measure of Old-Age Poverty

Abstract

ABSTRACT Traditional “head count” measures of poverty at advanced older ages understate the risk of falling into poverty because of survivorship bias due to the income-mortality gradient, which indicates that people in poverty have higher mortality rates than people with higher income. Survivorship bias is a form of sample selection bias. This paper presents a supplementary measure for poverty at older ages, based on an adaption of a model for correcting survivorship bias in rate of return data for mutual funds. Using U.S. longitudinal data from the Health and Retirement Study (HRS) for 1996 and 2002–2012 for the same cohort, we develop a new estimate of poverty at older ages that suggests that traditional cross-sectional measures understate the risk of falling into poverty by roughly a quarter. This finding has important implications for social programs that relate to the causes and consequences of the selectivity bias.