First, we demonstrated that states are as, if not more, important than counties in shaping the geographic variability in life expectancy in the US. Yet prior studies have largely focused on describing the inequality across counties16,17,19 and persistent clustering of high and low mortality counties20. In doing so, such studies have implicitly suggested that research and policy efforts should focus on the county-level processes and causes that might be the only drivers of longevity and premature mortality. We found that while counties accounted for 85% and 79% of the total variability in life expectancy for men and women, respectively, they accounted for less than 40% when states and counties were simultaneously modeled. This suggests that prior literature has considerably overestimated the importance of counties by omitting states. When geographic processes are likely to occur at multiple scales, empirical assessments should expand the units of analysis to accurately understand the scale at which action lies.
Second, there is a tendency – for no obvious reason that we are aware (except to consider geographic aggregations as a “proxy” for individuals) – to assume that a finer resolution of geographic aggregation (e.g., counties) is more important than a coarser resolution (e.g., states). However, we found that after accounting for counties, almost 50% of the total variation in life expectancy for men and over 40% for women were attributable to states. In fact, literature supports that processes at both state and county levels independently and simultaneously drive patterns of longevity and premature mortality.They produce a set of maps (see below) that show what difference it makes if we ignore either the state or the county level in our calculation of life expectancy. Professor Subramanian is, in my opinion, the world's foremost authority on multi-level analysis, so we need to pay close attention to these results.