Given that poor individuals face worse survival conditions than non-poor individuals, one can expect that a steeper income gradient in mortality leads, through stronger income-based selection, to a lower poverty rate at the old age (i.e. the "missing poor" hypothesis). This paper uses U.S. state-level data on poverty at age 65+ and life expectancy by income levels to provide an empirical test of the missing poor hypothesis. Using average temperature as an instrument for mortality differentials, we show that instrumented changes in mortality differentials have a negative and statistically significant effect on old-age poverty: a 1 % increase in the mortality differential implies a 16 % decrease in the 65+ headcount poverty rate. Using those regression results, we compute hypothetical old-age poverty rates while neutralizing the impact of the income gradient in mortality, and show that correcting for heterogeneity in income-based selection effects modifies the comparison of old-age poverty prevalence across states.