Empirically relevant critical values for hypothesis tests: A bootstrap approach