Chanting mortality rates among less educated Americans are difficult to interpret because the least educated groups (e.g. dropouts) become smaller and more negatively selected over time. New partial identification methods let us calculate mortality changes at constant education percentiles from 1992–2015. We find that middle-age mortality increases among non-Hispanic whites are driven almost entirely by changes in the bottom 10% of the education distribution. Drivers of mortality change differ substantially across groups. Deaths of despair explain a large share of mortality change among young non-Hispanic whites, but a small share among older whites and almost none among non-Hispanic blacks.