Data analyses usually make assumptions (which may be explicit or, more commonly, implicit): for example, “no unmeasured confounding”. When assumptions are untestable their potential importance can ...