A sensitivity analysis considers departures from randomized treatment assignment of various magnitudes. In an I × J block design, this means that the probability θij that individual j receives treatment in block i may depart from \(\overline {\theta }_{ij}=1/J\) . The sensitivity analysis determines the degree to which an inference about causal effects could change in the presence of departures from randomization, θij ≠ 1∕J of various magnitudes, thereby placing bounds on inference quantities, such as P-values, point estimates, and endpoints of confidence intervals. A sensitivity analysis replaces the true but useless statement “association does not imply causation,” by the equally true but far more useful statement “to explain the association actually seen in data, the bias in treatment assignment must exceed a particular magnitude.”