Is life expectancy really falling for low SES groups? Lagged selection bias and artifactual trends in mortality

Abstract Recent public health studies have made headlines by reporting that for some disadvantaged subpopulations in the U.S, recent period mortality rates are higher and life expectancy lower today compared to period rates from the past. In misleading syntax, this pattern has been described as a “fall” or “decline” in life expectancy. We illustrate how this important syntactic choice is akin to comparing average temperatures for the entire U.S in 1990 to only those for Alaska in 2010, and describing the difference as a “decline.” We explicate dynamics underlying what we call “Lagged Selection Bias” (LSB), and illustrate why comparing the same measure from two different periods in the presence of LSB does not actually identify a trend. If there are large and stable social disparities in health, 2010 life expectancy in a more disadvantaged group can be lower than 1990 life expectancy—even though everyone lives longer in the later period. We argue that it is long past time to examine an alternative—and arguably more plausible—interpretation of patterns in period mortality and period life expectancy. We urge more explicit attention to LSB and its dynamics among researchers interpreting changes in period measures stratified by subgroups whose composition is changing, and discuss approaches that have been used to manage it.