Researchers at Kaiser in Northern California wanted to evaluate the effect of use of their mail-order pharmacy service on adherence to medications. Some confounder information was available from...

Researchers at Kaiser in Northern California wanted to evaluate the effect of use of their mail-order pharmacy service on adherence to medications. Some confounder information was available from administrative databases, including age, sex, race/ethnicity, smoking, depression, and other co-morbidities, and whether the medication was covered by insurance, but there was concern about unmeasured confounders. Accordingly, they considered distance from the nearest brick-and-mortar Kaiser pharmacy to each member’s residence as an instrument. Consider this potential instrument in terms of its association with mail-order use, unconfoundedness, and possible indirect effects on the outcome not mediated by mail order use. What, if anything, could we do statistically to assess these assumptions?