R-squared obviously isn't the end-all be-all, but for that study he cites, they got .002 for the casualty rate. Also by their own results the racial makeup, ruralness, and education levels of a county were each larger factors in the 2012->2016 vote shift.
Imagine the need to exculpate Trump voters being so strong that you wind up doing all these mental gymnastics. Seems exhausting.