Definition: A statistic $T(X)$ is sufficient for $\theta$ if the conditional distribution $X \mid T(X)$ does not depend on $\theta$.